meeting-assistant/MeetingAssistant/Speakers/AzureSpeechSpeakerIdentityDiarizationClient.cs

using System.Threading.Channels;
using MeetingAssistant.Transcription;
using Microsoft.CognitiveServices.Speech;
using Microsoft.CognitiveServices.Speech.Audio;
using Microsoft.CognitiveServices.Speech.Transcription;
using Microsoft.Extensions.Options;

namespace MeetingAssistant.Speakers;

public sealed class AzureSpeechSpeakerIdentityDiarizationClient : ISpeakerIdentityDiarizationClient
{
    private readonly AzureSpeechOptions options;
    private readonly ILogger<AzureSpeechSpeakerIdentityDiarizationClient> logger;

    public AzureSpeechSpeakerIdentityDiarizationClient(
        IOptions<MeetingAssistantOptions> options,
        ILogger<AzureSpeechSpeakerIdentityDiarizationClient> logger)
    {
        this.options = options.Value.SpeakerIdentification.AzureSpeech;
        this.logger = logger;
    }

    public async Task<IReadOnlyList<TranscriptionSegment>> DiarizeAsync(
        string wavPath,
        CancellationToken cancellationToken)
    {
        var key = AzureSpeechStreamingTranscriptionProvider.ResolveKey(options);
        if (string.IsNullOrWhiteSpace(key))
        {
            logger.LogWarning(
                "Speaker identity Azure Speech matching is enabled but no key is configured directly or via {KeyEnv}",
                options.KeyEnv);
            return [];
        }

        var speechConfig = string.IsNullOrWhiteSpace(options.Region)
            ? SpeechConfig.FromEndpoint(new Uri(options.Endpoint), key)
            : SpeechConfig.FromSubscription(key, options.Region);
        speechConfig.SpeechRecognitionLanguage = ResolveRecognitionLanguage(options);
        speechConfig.OutputFormat = OutputFormat.Detailed;
        speechConfig.SetProperty(
            PropertyId.SpeechServiceResponse_DiarizeIntermediateResults,
            options.DiarizeIntermediateResults ? "true" : "false");

        using var audioConfig = AudioConfig.FromWavFileInput(wavPath);
        using var recognizer = new ConversationTranscriber(speechConfig, audioConfig);
        var channel = Channel.CreateUnbounded<TranscriptionSegment>();

        recognizer.Transcribed += (_, args) =>
        {
            if (args.Result.Reason != ResultReason.RecognizedSpeech)
            {
                return;
            }

            var start = TimeSpan.FromTicks(args.Result.OffsetInTicks);
            channel.Writer.TryWrite(new TranscriptionSegment(
                start,
                start + args.Result.Duration,
                NormalizeSpeakerId(args.Result.SpeakerId),
                args.Result.Text.Trim()));
        };
        recognizer.Canceled += (_, args) =>
        {
            if (args.Reason == CancellationReason.Error)
            {
                channel.Writer.TryComplete(new InvalidOperationException(
                    $"Azure Speech identity matching canceled: {args.ErrorCode} {args.ErrorDetails}"));
                return;
            }

            channel.Writer.TryComplete();
        };
        recognizer.SessionStopped += (_, _) => channel.Writer.TryComplete();

        await recognizer.StartTranscribingAsync();
        var segments = new List<TranscriptionSegment>();
        try
        {
            await foreach (var segment in channel.Reader.ReadAllAsync(cancellationToken))
            {
                segments.Add(segment);
            }
        }
        finally
        {
            try
            {
                var stopTask = recognizer.StopTranscribingAsync();
                if (options.RecognitionStopTimeout > TimeSpan.Zero)
                {
                    await stopTask.WaitAsync(options.RecognitionStopTimeout, CancellationToken.None);
                }
                else
                {
                    await stopTask.WaitAsync(CancellationToken.None);
                }
            }
            catch (TimeoutException)
            {
                logger.LogWarning("Timed out while stopping Azure Speech identity diarization recognizer");
            }
        }

        return segments;
    }

    internal static string ResolveRecognitionLanguage(AzureSpeechOptions options)
    {
        if (!options.Language.Equals("auto", StringComparison.OrdinalIgnoreCase))
        {
            return options.Language;
        }

        return options.AutoDetectLanguages
            .Select(language => language.Trim())
            .FirstOrDefault(language => !string.IsNullOrWhiteSpace(language))
            ?? "de-DE";
    }

    private static string NormalizeSpeakerId(string? speakerId)
    {
        return string.IsNullOrWhiteSpace(speakerId) || speakerId.Equals("Unknown", StringComparison.OrdinalIgnoreCase)
            ? "Unknown"
            : speakerId;
    }
}