Files
meeting-assistant/MeetingAssistant.Tests/RecordingCoordinatorTests.cs
T

1583 lines
65 KiB
C#

using MeetingAssistant.Recording;
using MeetingAssistant.Speakers;
using MeetingAssistant.Transcription;
using MeetingAssistant.MeetingNotes;
using MeetingAssistant.Summary;
using Microsoft.Extensions.Logging.Abstractions;
using Microsoft.Extensions.Options;
using System.Threading.Channels;
namespace MeetingAssistant.Tests;
public sealed class RecordingCoordinatorTests
{
private static async Task WaitUntilAsync(Func<bool> condition)
{
var deadline = DateTimeOffset.UtcNow.AddSeconds(5);
while (DateTimeOffset.UtcNow < deadline)
{
if (condition())
{
return;
}
await Task.Delay(25);
}
throw new TimeoutException("Condition was not met.");
}
[Fact]
public async Task ToggleStartsStreamingTranscriptionAndSecondToggleStopsIt()
{
var audioSource = new ControlledAudioSource();
var provider = new EchoStreamingTranscriptionProvider();
var store = new InMemoryTranscriptStore();
var noteStore = new InMemoryMeetingNoteStore();
var noteOpener = new CapturingMeetingNoteOpener();
var artifactStore = new InMemoryMeetingArtifactStore();
var audioArchive = new InMemoryRecordedAudioStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(provider),
store,
noteStore,
noteOpener,
artifactStore,
audioArchive,
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance);
var started = await coordinator.ToggleAsync(CancellationToken.None);
await audioSource.WriteAsync(new AudioChunk(new byte[] { 1, 0 }, 16000, 1), CancellationToken.None);
await store.WaitForTextAsync("chunk:2");
var stopped = await coordinator.ToggleAsync(CancellationToken.None);
Assert.True(started.IsRecording);
Assert.Equal(noteStore.SavedNote?.Path, noteOpener.OpenedPath);
Assert.Equal(started.MeetingNotePath, noteOpener.OpenedPath);
Assert.Equal(started.MeetingNotePath, artifactStore.CreatedArtifacts?.MeetingNotePath);
Assert.True(provider.FirstChunkWasObservedBeforeSourceCompleted);
Assert.False(stopped.IsRecording);
}
[Fact]
public async Task StartCreatesMeetingNoteLinkedToTranscriptAndOpensIt()
{
var audioSource = new ControlledAudioSource();
var transcriptStore = new InMemoryTranscriptStore("C:\\Vault\\Meetings\\Transcripts\\20260519-transcript.md");
var noteStore = new InMemoryMeetingNoteStore("C:\\Vault\\Meetings\\Notes\\20260519-meeting.md");
var noteOpener = new CapturingMeetingNoteOpener();
var artifactStore = new InMemoryMeetingArtifactStore();
var audioArchive = new InMemoryRecordedAudioStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new EchoStreamingTranscriptionProvider()),
transcriptStore,
noteStore,
noteOpener,
artifactStore,
audioArchive,
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions
{
Vault = new VaultOptions
{
AssistantContextFolder = "C:\\Vault\\Meetings\\Assistant Context",
SummariesFolder = "C:\\Vault\\Meetings\\Summaries"
}
}),
NullLogger<MeetingRecordingCoordinator>.Instance);
var status = await coordinator.StartAsync(CancellationToken.None);
Assert.True(status.IsRecording);
Assert.Equal("C:\\Vault\\Meetings\\Notes\\20260519-meeting.md", status.MeetingNotePath);
Assert.Equal("C:\\Vault\\Meetings\\Transcripts\\20260519-transcript.md", noteStore.SavedNote?.Frontmatter.Transcript);
Assert.StartsWith("C:\\Vault\\Meetings\\Assistant Context\\", noteStore.SavedNote?.Frontmatter.AssistantContext, StringComparison.Ordinal);
Assert.EndsWith("-assistant-context.md", noteStore.SavedNote?.Frontmatter.AssistantContext, StringComparison.Ordinal);
Assert.StartsWith("C:\\Vault\\Meetings\\Summaries\\", noteStore.SavedNote?.Frontmatter.Summary, StringComparison.Ordinal);
Assert.EndsWith("-summary.md", noteStore.SavedNote?.Frontmatter.Summary, StringComparison.Ordinal);
Assert.Equal("C:\\Vault\\Meetings\\Notes\\20260519-meeting.md", noteOpener.OpenedPath);
Assert.Equal("C:\\Vault\\Meetings\\Notes\\20260519-meeting.md", artifactStore.CreatedArtifacts?.MeetingNotePath);
Assert.Equal("C:\\Vault\\Meetings\\Transcripts\\20260519-transcript.md", artifactStore.CreatedArtifacts?.TranscriptPath);
Assert.StartsWith("C:\\Vault\\Meetings\\Assistant Context\\", artifactStore.CreatedArtifacts?.AssistantContextPath, StringComparison.Ordinal);
Assert.StartsWith("C:\\Vault\\Meetings\\Summaries\\", artifactStore.CreatedArtifacts?.SummaryPath, StringComparison.Ordinal);
await coordinator.StopAsync(CancellationToken.None);
}
[Fact]
public async Task StartUsesCurrentOutlookMeetingMetadataWhenAvailable()
{
var audioSource = new ControlledAudioSource();
var noteStore = new InMemoryMeetingNoteStore("C:\\Vault\\Meetings\\Notes\\metadata-meeting.md");
var artifactStore = new InMemoryMeetingArtifactStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new EchoStreamingTranscriptionProvider()),
new InMemoryTranscriptStore(),
noteStore,
new CapturingMeetingNoteOpener(),
artifactStore,
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance,
new FixedMeetingMetadataProvider(new MeetingMetadata(
"Architecture Sync",
["Ada <ada@example.com>", "Grace"],
"Review API shape",
DateTimeOffset.Parse("2026-05-19T11:00:00+02:00"))));
await coordinator.StartAsync(CancellationToken.None);
await WaitUntilAsync(() => artifactStore.Agenda == "Review API shape");
Assert.Equal("Architecture Sync", noteStore.SavedNote?.Frontmatter.Title);
Assert.Equal(["Ada <ada@example.com>", "Grace"], noteStore.SavedNote?.Frontmatter.Attendees);
Assert.Equal("Review API shape", artifactStore.Agenda);
Assert.Equal(DateTimeOffset.Parse("2026-05-19T11:00:00+02:00"), artifactStore.ScheduledEnd);
await coordinator.StopAsync(CancellationToken.None);
}
[Fact]
public async Task StartDoesNotWaitForSlowMeetingMetadataButAppliesItLater()
{
var audioSource = new ControlledAudioSource();
var noteStore = new InMemoryMeetingNoteStore("C:\\Vault\\Meetings\\Notes\\metadata-meeting.md");
var artifactStore = new InMemoryMeetingArtifactStore();
var provider = new BlockingMeetingMetadataProvider(new MeetingMetadata(
"Late Calendar Match",
["Ada"],
"Late agenda",
DateTimeOffset.Parse("2026-05-19T12:00:00+02:00")));
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new EchoStreamingTranscriptionProvider()),
new InMemoryTranscriptStore(),
noteStore,
new CapturingMeetingNoteOpener(),
artifactStore,
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance,
provider);
await coordinator.StartAsync(CancellationToken.None);
Assert.StartsWith("Meeting ", noteStore.SavedNote?.Frontmatter.Title, StringComparison.Ordinal);
provider.Release();
await WaitUntilAsync(() =>
noteStore.SavedNote?.Frontmatter.Title == "Late Calendar Match" &&
artifactStore.Agenda == "Late agenda");
Assert.Equal(["Ada"], noteStore.SavedNote?.Frontmatter.Attendees);
Assert.Equal("Late agenda", artifactStore.Agenda);
Assert.Equal(DateTimeOffset.Parse("2026-05-19T12:00:00+02:00"), artifactStore.ScheduledEnd);
await coordinator.StopAsync(CancellationToken.None);
}
[Fact]
public async Task StartCreatesMeetingNoteWhenMetadataProviderFails()
{
var audioSource = new ControlledAudioSource();
var noteStore = new InMemoryMeetingNoteStore("C:\\Vault\\Meetings\\Notes\\fallback-meeting.md");
var artifactStore = new InMemoryMeetingArtifactStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new EchoStreamingTranscriptionProvider()),
new InMemoryTranscriptStore("C:\\Vault\\Meetings\\Transcripts\\fallback-transcript.md"),
noteStore,
new CapturingMeetingNoteOpener(),
artifactStore,
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance,
new ThrowingMeetingMetadataProvider());
var status = await coordinator.StartAsync(CancellationToken.None);
Assert.True(status.IsRecording);
Assert.Equal("C:\\Vault\\Meetings\\Notes\\fallback-meeting.md", status.MeetingNotePath);
Assert.NotNull(artifactStore.CreatedArtifacts);
Assert.StartsWith("Meeting ", noteStore.SavedNote?.Frontmatter.Title, StringComparison.Ordinal);
await coordinator.StopAsync(CancellationToken.None);
}
[Fact]
public async Task StartContinuesWhenObsidianOpenFails()
{
var audioSource = new ControlledAudioSource();
var noteStore = new InMemoryMeetingNoteStore("C:\\Vault\\Meetings\\Notes\\open-fails.md");
var artifactStore = new InMemoryMeetingArtifactStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new EchoStreamingTranscriptionProvider()),
new InMemoryTranscriptStore("C:\\Vault\\Meetings\\Transcripts\\open-fails-transcript.md"),
noteStore,
new ThrowingMeetingNoteOpener(),
artifactStore,
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance);
var status = await coordinator.StartAsync(CancellationToken.None);
Assert.True(status.IsRecording);
Assert.Equal("C:\\Vault\\Meetings\\Notes\\open-fails.md", status.MeetingNotePath);
Assert.NotNull(artifactStore.CreatedArtifacts);
await coordinator.StopAsync(CancellationToken.None);
}
[Fact]
public async Task StopCompletesAudioCaptureAndDrainsFinalTranscriptionWindow()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk(new byte[] { 1, 0 }, 16000, 1));
var provider = new FinalSegmentOnAudioCompletionProvider();
var store = new InMemoryTranscriptStore();
var noteStore = new InMemoryMeetingNoteStore();
var noteOpener = new CapturingMeetingNoteOpener();
var artifactStore = new InMemoryMeetingArtifactStore();
var audioArchive = new InMemoryRecordedAudioStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(provider),
store,
noteStore,
noteOpener,
artifactStore,
audioArchive,
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WaitUntilCapturedAsync();
var stopped = await coordinator.StopAsync(CancellationToken.None);
Assert.False(stopped.IsRecording);
await store.WaitForTextAsync("final:2");
}
[Fact]
public async Task CaptureStartsEvenWhenTranscriptionProviderIsStillWarmingUp()
{
var audioSource = new ControlledAudioSource();
var provider = new BlockingBeforeTranscriptionProvider();
var store = new InMemoryTranscriptStore();
var audioArchive = new InMemoryRecordedAudioStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(provider),
store,
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
audioArchive,
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
await provider.WaitUntilWaitingForBackendAsync();
await audioSource.WriteAsync(new AudioChunk(new byte[] { 1, 0, 2, 0 }, 16000, 1), CancellationToken.None);
await audioArchive.WaitForAppendAsync();
Assert.Equal([4], audioArchive.AppendedChunkSizes);
provider.MarkBackendReady();
await store.WaitForTextAsync("chunk:4");
await coordinator.StopAsync(CancellationToken.None);
}
[Fact]
public async Task StartPassesDictationWordsToSpeechPipeline()
{
var audioSource = new ControlledAudioSource();
var provider = new CapturingPipelineOptionsProvider();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(provider),
new InMemoryTranscriptStore(),
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance,
dictationWordStore: new FixedDictationWordStore(["PBI", "Product Backlog Item"]));
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WriteAsync(new AudioChunk([1, 0], 16000, 1), CancellationToken.None);
await provider.OptionsObserved.Task.WaitAsync(TimeSpan.FromSeconds(5));
Assert.Equal(["PBI", "Product Backlog Item"], provider.Options?.DictationWords);
await coordinator.StopAsync(CancellationToken.None);
}
[Fact]
public async Task StopRewritesTranscriptWithFinalDiarizedSegmentsWhenAvailable()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk(new byte[] { 1, 0, 2, 0 }, 16000, 1));
var transcriptStore = new InMemoryTranscriptStore();
var finalizer = new CapturingTranscriptFinalizer(
[
new TranscriptionSegment(TimeSpan.Zero, TimeSpan.FromSeconds(1), "Speaker 0", "hello"),
new TranscriptionSegment(TimeSpan.FromSeconds(1), TimeSpan.FromSeconds(2), "Speaker 1", "there")
]);
var audioArchive = new InMemoryRecordedAudioStore("memory-recording.wav");
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider(), finalizer.FinalizeAsync),
transcriptStore,
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
audioArchive,
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
Assert.Equal("memory-recording.wav", finalizer.AudioPath);
Assert.Collection(
finalizer.LiveSegments,
segment => Assert.Equal("final:4", segment.Text));
Assert.Equal([4], audioArchive.AppendedChunkSizes);
Assert.True(audioArchive.Completed);
Assert.True(audioArchive.Deleted);
Assert.Collection(
transcriptStore.ReplacedSegments,
first =>
{
Assert.Equal("Speaker 0", first.Speaker);
Assert.Equal("hello", first.Text);
},
second =>
{
Assert.Equal("Speaker 1", second.Speaker);
Assert.Equal("there", second.Text);
});
}
[Fact]
public async Task StopRelabelsFinishedTranscriptWithLearnedSpeakerIdentity()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk(new byte[] { 1, 0, 2, 0 }, 16000, 1));
var transcriptStore = new InMemoryTranscriptStore();
var noteStore = new InMemoryMeetingNoteStore();
noteStore.UpdateSavedNote(MeetingNoteTemplate.Create(
"Identity Test",
DateTimeOffset.Parse("2026-05-20T12:00:00+02:00"),
attendees: ["Chris"],
transcriptPath: "memory-transcript.md",
assistantContextPath: "memory-context.md",
summaryPath: "memory-summary.md"));
var finalizer = new CapturingTranscriptFinalizer(
[
new TranscriptionSegment(TimeSpan.Zero, TimeSpan.FromSeconds(1), "Guest03", "hello")
]);
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider(), finalizer.FinalizeAsync),
transcriptStore,
noteStore,
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance,
speakerIdentificationService: new FixedSpeakerIdentificationService("Guest03", "Chris"));
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
Assert.Equal("Chris", transcriptStore.ReplacedSegments.Single().Speaker);
}
[Fact]
public async Task StopAddsIdentifiedSpeakerToMeetingAttendees()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk(new byte[] { 1, 0, 2, 0 }, 16000, 1));
var noteStore = new InMemoryMeetingNoteStore();
var finalizer = new CapturingTranscriptFinalizer(
[
new TranscriptionSegment(TimeSpan.Zero, TimeSpan.FromSeconds(1), "Guest03", "hello")
]);
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider(), finalizer.FinalizeAsync),
new InMemoryTranscriptStore(),
noteStore,
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance,
speakerIdentificationService: new FixedSpeakerIdentificationService(
"Guest03",
"Chris",
["Chris", "Christopher"]));
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
var attendees = Assert.IsType<List<string>>(noteStore.SavedNote?.Frontmatter.Attendees);
Assert.Contains("Chris", attendees);
}
[Fact]
public async Task StopDoesNotDuplicateIdentifiedSpeakerWhenAliasIsAlreadyAttendee()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk(new byte[] { 1, 0, 2, 0 }, 16000, 1));
var noteStore = new InMemoryMeetingNoteStore();
var finalizer = new CapturingTranscriptFinalizer(
[
new TranscriptionSegment(TimeSpan.Zero, TimeSpan.FromSeconds(1), "Guest03", "hello")
]);
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider(), finalizer.FinalizeAsync),
new InMemoryTranscriptStore(),
noteStore,
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance,
speakerIdentificationService: new FixedSpeakerIdentificationService(
"Guest03",
"Christopher",
["Christopher", "Chris"]));
await coordinator.StartAsync(CancellationToken.None);
noteStore.UpdateAttendees(["Chris <chris@example.com>"]);
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
var attendees = Assert.IsType<List<string>>(noteStore.SavedNote?.Frontmatter.Attendees);
Assert.Equal(["Chris <chris@example.com>"], attendees);
}
[Fact]
public async Task LiveSpeakerIdentityMatchRelabelsCurrentAndFutureTranscriptSegments()
{
var audioSource = new ControlledAudioSource();
var transcriptStore = new InMemoryTranscriptStore();
var speakerIdentification = new BlockingSpeakerIdentificationService("Guest03", "Chris");
var provider = new OrderedChunkProvider();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(provider),
transcriptStore,
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions
{
SpeakerIdentification = new SpeakerIdentificationOptions
{
InitialDelay = TimeSpan.Zero,
Interval = TimeSpan.FromMilliseconds(10)
}
}),
NullLogger<MeetingRecordingCoordinator>.Instance,
speakerIdentificationService: speakerIdentification);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WriteAsync(new AudioChunk(Samples(0, 1, 2, 3, 4, 5, 6, 7), 4, 1), CancellationToken.None);
await WaitUntilAsync(() => transcriptStore.Segments.Any(segment => segment.Text.Contains("first", StringComparison.Ordinal)));
await speakerIdentification.IdentificationObserved.Task.WaitAsync(TimeSpan.FromSeconds(5));
await WaitUntilAsync(() => transcriptStore.ReplacedSegments.Any(segment => segment.Text.Contains("first", StringComparison.Ordinal)));
Assert.Equal("Chris", transcriptStore.ReplacedSegments.Single().Speaker);
await Task.Delay(50);
await audioSource.WriteAsync(new AudioChunk(Samples(8, 9, 10, 11, 12, 13, 14, 15), 4, 1), CancellationToken.None);
await transcriptStore.WaitForTextAsync("second");
await coordinator.StopAsync(CancellationToken.None);
Assert.Collection(
transcriptStore.Segments,
first => Assert.Equal("Guest03", first.Speaker),
second => Assert.Equal("Chris", second.Speaker));
Assert.Contains(
transcriptStore.ReplacedSegments,
segment => segment.Text.Contains("first", StringComparison.Ordinal) && segment.Speaker == "Chris");
}
[Fact]
public async Task LiveSpeakerIdentityMatchAddsIdentifiedSpeakerToMeetingAttendees()
{
var audioSource = new ControlledAudioSource();
var noteStore = new InMemoryMeetingNoteStore();
var speakerIdentification = new BlockingSpeakerIdentificationService("Guest03", "Chris");
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new OrderedChunkProvider()),
new InMemoryTranscriptStore(),
noteStore,
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions
{
SpeakerIdentification = new SpeakerIdentificationOptions
{
InitialDelay = TimeSpan.Zero,
Interval = TimeSpan.FromMilliseconds(10)
}
}),
NullLogger<MeetingRecordingCoordinator>.Instance,
speakerIdentificationService: speakerIdentification);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WriteAsync(new AudioChunk(Samples(0, 1, 2, 3, 4, 5, 6, 7), 4, 1), CancellationToken.None);
await WaitUntilAsync(() => noteStore.SavedNote?.Frontmatter.Attendees.Contains("Chris") == true);
var attendees = Assert.IsType<List<string>>(noteStore.SavedNote?.Frontmatter.Attendees);
Assert.Contains("Chris", attendees);
await coordinator.StopAsync(CancellationToken.None);
}
[Fact]
public async Task LiveSpeakerIdentificationRetriesWhenAttendeesChange()
{
var audioSource = new ControlledAudioSource();
var noteStore = new InMemoryMeetingNoteStore();
var speakerIdentification = new CountingSpeakerIdentificationService();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new OrderedChunkProvider()),
new InMemoryTranscriptStore(),
noteStore,
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions
{
SpeakerIdentification = new SpeakerIdentificationOptions
{
InitialDelay = TimeSpan.Zero,
Interval = TimeSpan.FromMilliseconds(20)
}
}),
NullLogger<MeetingRecordingCoordinator>.Instance,
speakerIdentificationService: speakerIdentification);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WriteAsync(new AudioChunk(Samples(0, 1, 2, 3, 4, 5, 6, 7), 4, 1), CancellationToken.None);
await WaitUntilAsync(() => speakerIdentification.Requests.Count == 1);
noteStore.UpdateAttendees(["Chris"]);
await WaitUntilAsync(() => speakerIdentification.Requests.Count == 2);
await coordinator.StopAsync(CancellationToken.None);
Assert.Equal(["Chris"], speakerIdentification.Requests.Last().MeetingNote.Frontmatter.Attendees);
}
[Fact]
public async Task LiveSpeakerIdentificationRetriesWhenNewUnmappedSpeakerAppears()
{
var audioSource = new ControlledAudioSource();
var speakerIdentification = new CountingSpeakerIdentificationService();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new ChangingSpeakerOrderedChunkProvider()),
new InMemoryTranscriptStore(),
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions
{
SpeakerIdentification = new SpeakerIdentificationOptions
{
InitialDelay = TimeSpan.Zero,
Interval = TimeSpan.FromMilliseconds(20)
}
}),
NullLogger<MeetingRecordingCoordinator>.Instance,
speakerIdentificationService: speakerIdentification);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WriteAsync(new AudioChunk(Samples(0, 1, 2, 3, 4, 5, 6, 7), 4, 1), CancellationToken.None);
await WaitUntilAsync(() => speakerIdentification.Requests.Count == 1);
await audioSource.WriteAsync(new AudioChunk(Samples(8, 9, 10, 11, 12, 13, 14, 15), 4, 1), CancellationToken.None);
await WaitUntilAsync(() => speakerIdentification.Requests.Count == 2);
await coordinator.StopAsync(CancellationToken.None);
var samples = Assert.IsAssignableFrom<IReadOnlyList<SpeakerAudioSample>>(
speakerIdentification.Requests.Last().Samples);
Assert.Contains(samples, sample => sample.Speaker == "Guest04");
}
[Theory]
[InlineData(0, null)]
[InlineData(1, null)]
[InlineData(2, 2)]
[InlineData(5, 5)]
public async Task StopUsesMeetingNoteAttendeeCountAsSpeakerHintWhenThereAreMultipleAttendees(
int attendeeCount,
int? expectedNumSpeakers)
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk([1, 0, 2, 0], 16000, 1));
var noteStore = new InMemoryMeetingNoteStore();
var finalizer = new CapturingTranscriptFinalizer(
[
new TranscriptionSegment(TimeSpan.Zero, TimeSpan.FromSeconds(1), "Speaker 0", "hello")
]);
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider(), finalizer.FinalizeAsync),
new InMemoryTranscriptStore(),
noteStore,
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
noteStore.UpdateAttendees(Enumerable.Range(1, attendeeCount).Select(index => $"Person {index}"));
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
Assert.Equal(expectedNumSpeakers, finalizer.Options?.NumSpeakers);
}
[Fact]
public async Task StopAddsMeetingEndTimeAndRunsSummaryAfterFinishedTranscript()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk([1, 0, 2, 0], 16000, 1));
var noteStore = new InMemoryMeetingNoteStore();
var artifactStore = new InMemoryMeetingArtifactStore();
var summaryPipeline = new CapturingMeetingSummaryPipeline();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider()),
new InMemoryTranscriptStore(),
noteStore,
new CapturingMeetingNoteOpener(),
artifactStore,
new InMemoryRecordedAudioStore(),
summaryPipeline,
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
var startTime = noteStore.SavedNote?.Frontmatter.StartTime;
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
Assert.NotNull(startTime);
Assert.NotNull(noteStore.SavedNote?.Frontmatter.EndTime);
Assert.True(noteStore.SavedNote?.Frontmatter.EndTime >= startTime);
Assert.Equal(artifactStore.CreatedArtifacts, summaryPipeline.Artifacts);
}
[Fact]
public async Task StopUpdatesTranscriptMetadataWithMeetingEndTime()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk([1, 0, 2, 0], 16000, 1));
var transcriptStore = new InMemoryTranscriptStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider()),
transcriptStore,
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
new InMemoryMeetingArtifactStore(),
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions()),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
Assert.NotNull(transcriptStore.MetadataMeetingNote?.Frontmatter.EndTime);
}
[Fact]
public async Task StopUpdatesAssistantContextStateThroughSummaryLifecycle()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk([1, 0, 2, 0], 16000, 1));
var artifactStore = new InMemoryMeetingArtifactStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider()),
new InMemoryTranscriptStore(),
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
artifactStore,
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(CreateOptionsWithoutFinalizer()),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
Assert.Equal(
[AssistantContextState.Summarizing, AssistantContextState.Finished],
artifactStore.States);
}
[Fact]
public async Task StopMarksAssistantContextErrorWhenSummaryFails()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk([1, 0, 2, 0], 16000, 1));
var artifactStore = new InMemoryMeetingArtifactStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider()),
new InMemoryTranscriptStore(),
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
artifactStore,
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(succeeded: false),
Options.Create(CreateOptionsWithoutFinalizer()),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
Assert.Equal(
[AssistantContextState.Summarizing, AssistantContextState.Error],
artifactStore.States);
}
[Fact]
public async Task StopMarksSpeakerRecognitionBeforeSummaryWhenFinalizerIsConfigured()
{
var audioSource = new CapturedChunkThenCancelAudioSource(new AudioChunk([1, 0, 2, 0], 16000, 1));
var artifactStore = new InMemoryMeetingArtifactStore();
var coordinator = new MeetingRecordingCoordinator(
audioSource,
new TestSpeechRecognitionPipelineFactory(new FinalSegmentOnAudioCompletionProvider()),
new InMemoryTranscriptStore(),
new InMemoryMeetingNoteStore(),
new CapturingMeetingNoteOpener(),
artifactStore,
new InMemoryRecordedAudioStore(),
new CapturingMeetingSummaryPipeline(),
Options.Create(new MeetingAssistantOptions
{
Recording = new RecordingOptions { TranscriptionProvider = "whisper-local" },
WhisperLocal = new WhisperLocalOptions
{
Diarization = new PyannoteDiarizationOptions { Enabled = true }
}
}),
NullLogger<MeetingRecordingCoordinator>.Instance);
await coordinator.StartAsync(CancellationToken.None);
await audioSource.WaitUntilCapturedAsync();
await coordinator.StopAsync(CancellationToken.None);
Assert.Equal(
[AssistantContextState.SpeakerRecognition, AssistantContextState.Summarizing, AssistantContextState.Finished],
artifactStore.States);
}
[Fact]
public async Task VaultTranscriptStoreCreatesConfiguredFolderAndAppendsSegments()
{
var vaultFolder = Path.Combine(Path.GetTempPath(), "meeting-assistant-tests", Guid.NewGuid().ToString("N"));
var store = new VaultTranscriptStore(
Options.Create(new MeetingAssistantOptions
{
Vault = new VaultOptions { TranscriptsFolder = vaultFolder }
}),
NullLogger<VaultTranscriptStore>.Instance);
var session = await store.CreateSessionAsync(CancellationToken.None);
await store.AppendAsync(
session,
new TranscriptionSegment(TimeSpan.Zero, TimeSpan.FromSeconds(1), "Unknown", "hello vault"),
CancellationToken.None);
Assert.True(Directory.Exists(vaultFolder));
Assert.EndsWith(".md", session.TranscriptPath, StringComparison.Ordinal);
Assert.Contains("hello vault", await File.ReadAllTextAsync(session.TranscriptPath));
}
[Fact]
public async Task TemporaryRecordedAudioStoreCreatesConfiguredFolderAndWritesPcmWav()
{
var recordingFolder = Path.Combine(Path.GetTempPath(), "meeting-assistant-tests", Guid.NewGuid().ToString("N"));
var store = new TemporaryRecordedAudioStore(
Options.Create(new MeetingAssistantOptions
{
Recording = new RecordingOptions
{
SampleRate = 16000,
Channels = 1,
TemporaryRecordingsFolder = recordingFolder
}
}),
NullLogger<TemporaryRecordedAudioStore>.Instance);
await using var session = await store.CreateSessionAsync(CancellationToken.None);
await session.AppendAsync(new AudioChunk([1, 0, 2, 0], 16000, 1), CancellationToken.None);
await session.CompleteAsync(CancellationToken.None);
Assert.True(Directory.Exists(recordingFolder));
Assert.EndsWith(".wav", session.AudioPath, StringComparison.Ordinal);
Assert.True(new FileInfo(session.AudioPath).Length > 44);
}
[Fact]
public async Task TemporaryRecordedAudioStoreDeletesStaleRecordingsOnStartup()
{
var recordingFolder = Path.Combine(Path.GetTempPath(), "meeting-assistant-tests", Guid.NewGuid().ToString("N"));
Directory.CreateDirectory(recordingFolder);
var staleRecording = Path.Combine(recordingFolder, "stale.wav");
var unrelatedFile = Path.Combine(recordingFolder, "keep.txt");
await File.WriteAllTextAsync(staleRecording, "stale");
await File.WriteAllTextAsync(unrelatedFile, "keep");
var store = new TemporaryRecordedAudioStore(
Options.Create(new MeetingAssistantOptions
{
Recording = new RecordingOptions { TemporaryRecordingsFolder = recordingFolder }
}),
NullLogger<TemporaryRecordedAudioStore>.Instance);
await store.DeleteStaleRecordingsAsync(CancellationToken.None);
Assert.False(File.Exists(staleRecording));
Assert.True(File.Exists(unrelatedFile));
}
private static MeetingAssistantOptions CreateOptionsWithoutFinalizer()
{
return new MeetingAssistantOptions
{
Recording = new RecordingOptions { TranscriptionProvider = "whisper-local" },
WhisperLocal = new WhisperLocalOptions
{
Diarization = new PyannoteDiarizationOptions { Enabled = false }
}
};
}
private sealed class InMemoryTranscriptStore : ITranscriptStore
{
private readonly List<TranscriptionSegment> segments = [];
private readonly TaskCompletionSource segmentWritten = new(TaskCreationOptions.RunContinuationsAsynchronously);
private readonly string transcriptPath;
public InMemoryTranscriptStore(string transcriptPath = "memory-transcript.md")
{
this.transcriptPath = transcriptPath;
}
public Task<TranscriptSession> CreateSessionAsync(CancellationToken cancellationToken)
{
return Task.FromResult(new TranscriptSession(transcriptPath));
}
public Task AppendAsync(TranscriptSession session, TranscriptionSegment segment, CancellationToken cancellationToken)
{
segments.Add(segment);
segmentWritten.TrySetResult();
return Task.CompletedTask;
}
public async Task WaitForTextAsync(string text)
{
var deadline = DateTimeOffset.UtcNow.AddSeconds(5);
while (DateTimeOffset.UtcNow < deadline)
{
if (segments.Any(segment => segment.Text.Contains(text, StringComparison.Ordinal)))
{
return;
}
await segmentWritten.Task.WaitAsync(TimeSpan.FromMilliseconds(100));
}
throw new TimeoutException($"Segment containing '{text}' was not written.");
}
public IReadOnlyList<TranscriptionSegment> ReplacedSegments { get; private set; } = [];
public IReadOnlyList<TranscriptionSegment> Segments => segments;
public MeetingNote? MetadataMeetingNote { get; private set; }
public Task ReplaceAsync(
TranscriptSession session,
IReadOnlyList<TranscriptionSegment> replacementSegments,
CancellationToken cancellationToken)
{
ReplacedSegments = replacementSegments;
return Task.CompletedTask;
}
public Task UpdateMetadataAsync(
TranscriptSession session,
MeetingSessionArtifacts artifacts,
MeetingNote meetingNote,
CancellationToken cancellationToken)
{
MetadataMeetingNote = meetingNote;
return Task.CompletedTask;
}
}
private sealed class InMemoryMeetingNoteStore : IMeetingNoteStore
{
private readonly string notePath;
public InMemoryMeetingNoteStore(string notePath = "memory-meeting.md")
{
this.notePath = notePath;
}
public MeetingNote? SavedNote { get; private set; }
public Task<MeetingNote> SaveAsync(MeetingNote note, CancellationToken cancellationToken)
{
SavedNote = note with { Path = notePath };
return Task.FromResult(SavedNote);
}
public Task<MeetingNote> ReadAsync(string path, CancellationToken cancellationToken)
{
return Task.FromResult(SavedNote ?? throw new FileNotFoundException(path));
}
public void UpdateAttendees(IEnumerable<string> attendees)
{
if (SavedNote is null)
{
throw new InvalidOperationException("No meeting note has been saved.");
}
SavedNote.Frontmatter.Attendees = attendees.ToList();
}
public void UpdateSavedNote(MeetingNote note)
{
SavedNote = note with { Path = notePath };
}
}
private sealed class CapturingMeetingNoteOpener : IMeetingNoteOpener
{
public string? OpenedPath { get; private set; }
public Task OpenAsync(string notePath, CancellationToken cancellationToken)
{
OpenedPath = notePath;
return Task.CompletedTask;
}
}
private sealed class InMemoryMeetingArtifactStore : IMeetingArtifactStore
{
public MeetingSessionArtifacts? CreatedArtifacts { get; private set; }
public List<AssistantContextState> States { get; } = [];
public string? Agenda { get; private set; }
public DateTimeOffset? ScheduledEnd { get; private set; }
public Task CreateAssistantContextAsync(
MeetingSessionArtifacts artifacts,
string agenda,
DateTimeOffset? scheduledEnd,
CancellationToken cancellationToken)
{
CreatedArtifacts = artifacts;
Agenda = agenda;
ScheduledEnd = scheduledEnd;
return Task.CompletedTask;
}
public Task UpdateAssistantContextStateAsync(
MeetingSessionArtifacts artifacts,
AssistantContextState state,
CancellationToken cancellationToken)
{
States.Add(state);
return Task.CompletedTask;
}
public Task UpdateAssistantContextMetadataAsync(
MeetingSessionArtifacts artifacts,
string agenda,
DateTimeOffset? scheduledEnd,
CancellationToken cancellationToken)
{
Agenda = agenda;
ScheduledEnd = scheduledEnd;
return Task.CompletedTask;
}
}
private sealed class FixedMeetingMetadataProvider : IMeetingMetadataProvider
{
private readonly MeetingMetadata? metadata;
public FixedMeetingMetadataProvider(MeetingMetadata? metadata)
{
this.metadata = metadata;
}
public Task<MeetingMetadata?> GetCurrentMeetingAsync(
DateTimeOffset startedAt,
CancellationToken cancellationToken)
{
return Task.FromResult(metadata);
}
}
private sealed class ThrowingMeetingMetadataProvider : IMeetingMetadataProvider
{
public Task<MeetingMetadata?> GetCurrentMeetingAsync(
DateTimeOffset startedAt,
CancellationToken cancellationToken)
{
throw new InvalidOperationException("metadata unavailable");
}
}
private sealed class BlockingMeetingMetadataProvider : IMeetingMetadataProvider
{
private readonly MeetingMetadata metadata;
private readonly TaskCompletionSource release = new(TaskCreationOptions.RunContinuationsAsynchronously);
public BlockingMeetingMetadataProvider(MeetingMetadata metadata)
{
this.metadata = metadata;
}
public void Release()
{
release.TrySetResult();
}
public async Task<MeetingMetadata?> GetCurrentMeetingAsync(
DateTimeOffset startedAt,
CancellationToken cancellationToken)
{
await release.Task.WaitAsync(cancellationToken);
return metadata;
}
}
private sealed class ThrowingMeetingNoteOpener : IMeetingNoteOpener
{
public Task OpenAsync(string notePath, CancellationToken cancellationToken)
{
throw new InvalidOperationException("obsidian unavailable");
}
}
private sealed class CapturingMeetingSummaryPipeline : IMeetingSummaryPipeline
{
private readonly bool succeeded;
public CapturingMeetingSummaryPipeline(bool succeeded = true)
{
this.succeeded = succeeded;
}
public MeetingSessionArtifacts? Artifacts { get; private set; }
public Task<MeetingSummaryRunResult> RunAsync(
MeetingSessionArtifacts artifacts,
CancellationToken cancellationToken)
{
Artifacts = artifacts;
return Task.FromResult(new MeetingSummaryRunResult(
artifacts.SummaryPath,
succeeded ? "summary ok" : "summary failed",
succeeded,
succeeded ? null : "error"));
}
}
private sealed class InMemoryRecordedAudioStore : IRecordedAudioStore
{
private readonly string audioPath;
public InMemoryRecordedAudioStore(string audioPath = "memory-recording.wav")
{
this.audioPath = audioPath;
}
public List<int> AppendedChunkSizes { get; } = [];
public bool Completed { get; private set; }
public bool Deleted { get; private set; }
private TaskCompletionSource AppendObserved { get; set; } =
new(TaskCreationOptions.RunContinuationsAsynchronously);
public Task WaitForAppendAsync()
{
return AppendObserved.Task.WaitAsync(TimeSpan.FromSeconds(5));
}
public Task<IRecordedAudioSink> CreateSessionAsync(CancellationToken cancellationToken)
{
return Task.FromResult<IRecordedAudioSink>(new Sink(this, audioPath));
}
public Task DeleteStaleRecordingsAsync(CancellationToken cancellationToken)
{
return Task.CompletedTask;
}
private sealed class Sink : IRecordedAudioSink
{
private readonly InMemoryRecordedAudioStore store;
public Sink(InMemoryRecordedAudioStore store, string audioPath)
{
this.store = store;
AudioPath = audioPath;
}
public string AudioPath { get; }
public Task AppendAsync(AudioChunk chunk, CancellationToken cancellationToken)
{
store.AppendedChunkSizes.Add(chunk.Pcm.Length);
store.AppendObserved.TrySetResult();
return Task.CompletedTask;
}
public Task CompleteAsync(CancellationToken cancellationToken)
{
store.Completed = true;
return Task.CompletedTask;
}
public ValueTask DisposeAsync()
{
return ValueTask.CompletedTask;
}
public Task DeleteAsync(CancellationToken cancellationToken)
{
store.Deleted = true;
return Task.CompletedTask;
}
}
}
private sealed class TestSpeechRecognitionPipelineFactory : ISpeechRecognitionPipelineFactory
{
private readonly IStreamingTranscriptionProvider provider;
private readonly Func<string, IReadOnlyList<TranscriptionSegment>, SpeechRecognitionPipelineOptions, CancellationToken, Task<IReadOnlyList<TranscriptionSegment>>> finalize;
public TestSpeechRecognitionPipelineFactory(
IStreamingTranscriptionProvider provider,
Func<string, IReadOnlyList<TranscriptionSegment>, SpeechRecognitionPipelineOptions, CancellationToken, Task<IReadOnlyList<TranscriptionSegment>>>? finalize = null)
{
this.provider = provider;
this.finalize = finalize ?? ((_, _, _, _) => Task.FromResult<IReadOnlyList<TranscriptionSegment>>([]));
}
public ISpeechRecognitionPipeline Create()
{
return new TestSpeechRecognitionPipeline(provider, finalize);
}
}
private sealed class TestSpeechRecognitionPipeline : StreamingSpeechRecognitionPipeline
{
private readonly Func<string, IReadOnlyList<TranscriptionSegment>, SpeechRecognitionPipelineOptions, CancellationToken, Task<IReadOnlyList<TranscriptionSegment>>> finalize;
public TestSpeechRecognitionPipeline(
IStreamingTranscriptionProvider provider,
Func<string, IReadOnlyList<TranscriptionSegment>, SpeechRecognitionPipelineOptions, CancellationToken, Task<IReadOnlyList<TranscriptionSegment>>> finalize)
: base(provider)
{
this.finalize = finalize;
}
protected override Task<IReadOnlyList<TranscriptionSegment>> BuildFinishedTranscriptAsync(
string audioPath,
IReadOnlyList<TranscriptionSegment> liveSegments,
SpeechRecognitionPipelineOptions options,
CancellationToken cancellationToken)
{
return finalize(audioPath, liveSegments, options, cancellationToken);
}
}
private sealed class CapturingTranscriptFinalizer
{
private readonly IReadOnlyList<TranscriptionSegment> segments;
public CapturingTranscriptFinalizer(IReadOnlyList<TranscriptionSegment> segments)
{
this.segments = segments;
}
public string? AudioPath { get; private set; }
public IReadOnlyList<TranscriptionSegment> LiveSegments { get; private set; } = [];
public SpeechRecognitionPipelineOptions? Options { get; private set; }
public Task<IReadOnlyList<TranscriptionSegment>> FinalizeAsync(
string audioPath,
IReadOnlyList<TranscriptionSegment> liveSegments,
SpeechRecognitionPipelineOptions options,
CancellationToken cancellationToken)
{
AudioPath = audioPath;
LiveSegments = liveSegments;
Options = options;
return Task.FromResult(segments);
}
}
private sealed class FixedSpeakerIdentificationService : ISpeakerIdentificationService
{
private readonly string sourceSpeaker;
private readonly string targetSpeaker;
private readonly IReadOnlyList<string> acceptedNames;
public FixedSpeakerIdentificationService(
string sourceSpeaker,
string targetSpeaker,
IReadOnlyList<string>? acceptedNames = null)
{
this.sourceSpeaker = sourceSpeaker;
this.targetSpeaker = targetSpeaker;
this.acceptedNames = acceptedNames ?? [targetSpeaker];
}
public Task<SpeakerIdentificationResult> ProcessFinishedTranscriptAsync(
SpeakerIdentificationRequest request,
CancellationToken cancellationToken)
{
return Task.FromResult(new SpeakerIdentificationResult(
request.Segments.Select(segment => segment.Speaker == sourceSpeaker
? segment with { Speaker = targetSpeaker }
: segment).ToList(),
new Dictionary<string, string> { [sourceSpeaker] = targetSpeaker },
[new SpeakerIdentityAttendeeMatch(targetSpeaker, acceptedNames)]));
}
public Task<SpeakerIdentificationResult> IdentifyKnownSpeakersAsync(
SpeakerIdentificationRequest request,
CancellationToken cancellationToken)
{
return ProcessFinishedTranscriptAsync(request, cancellationToken);
}
}
private sealed class BlockingSpeakerIdentificationService : ISpeakerIdentificationService
{
private readonly string sourceSpeaker;
private readonly string targetSpeaker;
private readonly IReadOnlyList<string> acceptedNames;
public BlockingSpeakerIdentificationService(
string sourceSpeaker,
string targetSpeaker,
IReadOnlyList<string>? acceptedNames = null)
{
this.sourceSpeaker = sourceSpeaker;
this.targetSpeaker = targetSpeaker;
this.acceptedNames = acceptedNames ?? [targetSpeaker];
}
public TaskCompletionSource IdentificationObserved { get; } =
new(TaskCreationOptions.RunContinuationsAsynchronously);
public Task<SpeakerIdentificationResult> IdentifyKnownSpeakersAsync(
SpeakerIdentificationRequest request,
CancellationToken cancellationToken)
{
IdentificationObserved.TrySetResult();
return Task.FromResult(new SpeakerIdentificationResult(
request.Segments,
new Dictionary<string, string> { [sourceSpeaker] = targetSpeaker },
[new SpeakerIdentityAttendeeMatch(targetSpeaker, acceptedNames)]));
}
public Task<SpeakerIdentificationResult> ProcessFinishedTranscriptAsync(
SpeakerIdentificationRequest request,
CancellationToken cancellationToken)
{
return Task.FromResult(new SpeakerIdentificationResult(request.Segments, new Dictionary<string, string>()));
}
}
private sealed class CountingSpeakerIdentificationService : ISpeakerIdentificationService
{
public List<SpeakerIdentificationRequest> Requests { get; } = [];
public Task<SpeakerIdentificationResult> IdentifyKnownSpeakersAsync(
SpeakerIdentificationRequest request,
CancellationToken cancellationToken)
{
Requests.Add(request);
return Task.FromResult(new SpeakerIdentificationResult(request.Segments, new Dictionary<string, string>()));
}
public Task<SpeakerIdentificationResult> ProcessFinishedTranscriptAsync(
SpeakerIdentificationRequest request,
CancellationToken cancellationToken)
{
return Task.FromResult(new SpeakerIdentificationResult(request.Segments, new Dictionary<string, string>()));
}
}
private sealed class FixedDictationWordStore : IDictationWordStore
{
private readonly IReadOnlyList<string> words;
public FixedDictationWordStore(IReadOnlyList<string> words)
{
this.words = words;
}
public Task<IReadOnlyList<string>> ReadWordsAsync(CancellationToken cancellationToken)
{
return Task.FromResult(words);
}
public Task<IReadOnlyList<string>> AddWordAsync(string word, CancellationToken cancellationToken)
{
return Task.FromResult(words);
}
}
private sealed class CapturingPipelineOptionsProvider : IStreamingTranscriptionProvider
{
public TaskCompletionSource OptionsObserved { get; } =
new(TaskCreationOptions.RunContinuationsAsynchronously);
public SpeechRecognitionPipelineOptions? Options { get; private set; }
public async IAsyncEnumerable<TranscriptionSegment> TranscribeAsync(
IAsyncEnumerable<AudioChunk> audio,
SpeechRecognitionPipelineOptions options,
[System.Runtime.CompilerServices.EnumeratorCancellation] CancellationToken cancellationToken)
{
Options = options;
OptionsObserved.TrySetResult();
await foreach (var _ in audio.WithCancellation(cancellationToken))
{
yield return new TranscriptionSegment(TimeSpan.Zero, TimeSpan.Zero, "Unknown", "ok");
}
}
}
private sealed class OrderedChunkProvider : IStreamingTranscriptionProvider
{
public async IAsyncEnumerable<TranscriptionSegment> TranscribeAsync(
IAsyncEnumerable<AudioChunk> audio,
SpeechRecognitionPipelineOptions options,
[System.Runtime.CompilerServices.EnumeratorCancellation] CancellationToken cancellationToken)
{
var index = 0;
await foreach (var _ in audio.WithCancellation(cancellationToken))
{
var text = index++ == 0 ? "first" : "second";
yield return new TranscriptionSegment(
TimeSpan.FromSeconds(index - 1),
TimeSpan.FromSeconds(index + 1),
"Guest03",
$"{text} has enough words for identification.");
}
}
}
private sealed class ChangingSpeakerOrderedChunkProvider : IStreamingTranscriptionProvider
{
public async IAsyncEnumerable<TranscriptionSegment> TranscribeAsync(
IAsyncEnumerable<AudioChunk> audio,
SpeechRecognitionPipelineOptions options,
[System.Runtime.CompilerServices.EnumeratorCancellation] CancellationToken cancellationToken)
{
var index = 0;
await foreach (var _ in audio.WithCancellation(cancellationToken))
{
var speaker = index++ == 0 ? "Guest03" : "Guest04";
yield return new TranscriptionSegment(
TimeSpan.FromSeconds(index - 1),
TimeSpan.FromSeconds(index + 1),
speaker,
"This segment has enough words for identification.");
}
}
}
private sealed class ControlledAudioSource : IMeetingAudioSource
{
private readonly Channel<AudioChunk> chunks = Channel.CreateUnbounded<AudioChunk>();
public IAsyncEnumerable<AudioChunk> CaptureAsync(CancellationToken cancellationToken)
{
return chunks.Reader.ReadAllAsync(cancellationToken);
}
public ValueTask WriteAsync(AudioChunk chunk, CancellationToken cancellationToken)
{
return chunks.Writer.WriteAsync(chunk, cancellationToken);
}
}
private sealed class CapturedChunkThenCancelAudioSource : IMeetingAudioSource
{
private readonly AudioChunk chunk;
private readonly TaskCompletionSource captured = new(TaskCreationOptions.RunContinuationsAsynchronously);
public CapturedChunkThenCancelAudioSource(AudioChunk chunk)
{
this.chunk = chunk;
}
public async IAsyncEnumerable<AudioChunk> CaptureAsync(
[System.Runtime.CompilerServices.EnumeratorCancellation] CancellationToken cancellationToken)
{
yield return chunk;
captured.TrySetResult();
try
{
await Task.Delay(Timeout.InfiniteTimeSpan, cancellationToken);
}
finally
{
captured.TrySetResult();
}
}
public Task WaitUntilCapturedAsync()
{
return captured.Task.WaitAsync(TimeSpan.FromSeconds(5));
}
}
private sealed class EchoStreamingTranscriptionProvider : IStreamingTranscriptionProvider
{
public bool FirstChunkWasObservedBeforeSourceCompleted { get; private set; }
public async IAsyncEnumerable<TranscriptionSegment> TranscribeAsync(
IAsyncEnumerable<AudioChunk> audio,
SpeechRecognitionPipelineOptions options,
[System.Runtime.CompilerServices.EnumeratorCancellation] CancellationToken cancellationToken)
{
await foreach (var chunk in audio.WithCancellation(cancellationToken))
{
FirstChunkWasObservedBeforeSourceCompleted = true;
yield return new TranscriptionSegment(TimeSpan.Zero, TimeSpan.Zero, "Unknown", $"chunk:{chunk.Pcm.Length}");
}
}
}
private sealed class FinalSegmentOnAudioCompletionProvider : IStreamingTranscriptionProvider
{
public async IAsyncEnumerable<TranscriptionSegment> TranscribeAsync(
IAsyncEnumerable<AudioChunk> audio,
SpeechRecognitionPipelineOptions options,
[System.Runtime.CompilerServices.EnumeratorCancellation] CancellationToken cancellationToken)
{
var byteCount = 0;
await foreach (var chunk in audio.WithCancellation(cancellationToken))
{
byteCount += chunk.Pcm.Length;
}
yield return new TranscriptionSegment(TimeSpan.Zero, TimeSpan.Zero, "Unknown", $"final:{byteCount}");
}
}
private sealed class BlockingBeforeTranscriptionProvider : IStreamingTranscriptionProvider
{
private readonly TaskCompletionSource waitingForBackend =
new(TaskCreationOptions.RunContinuationsAsynchronously);
private readonly TaskCompletionSource backendReady =
new(TaskCreationOptions.RunContinuationsAsynchronously);
public Task WaitUntilWaitingForBackendAsync()
{
return waitingForBackend.Task.WaitAsync(TimeSpan.FromSeconds(5));
}
public void MarkBackendReady()
{
backendReady.TrySetResult();
}
public async IAsyncEnumerable<TranscriptionSegment> TranscribeAsync(
IAsyncEnumerable<AudioChunk> audio,
SpeechRecognitionPipelineOptions options,
[System.Runtime.CompilerServices.EnumeratorCancellation] CancellationToken cancellationToken)
{
waitingForBackend.TrySetResult();
await backendReady.Task.WaitAsync(cancellationToken);
await foreach (var chunk in audio.WithCancellation(cancellationToken))
{
yield return new TranscriptionSegment(TimeSpan.Zero, TimeSpan.Zero, "Unknown", $"chunk:{chunk.Pcm.Length}");
}
}
}
private static byte[] Samples(params short[] samples)
{
var bytes = new byte[samples.Length * sizeof(short)];
Buffer.BlockCopy(samples, 0, bytes, 0, bytes.Length);
return bytes;
}
}