openai/openai-dotnet
Publicmirrored from https://github.com/openai/openai-dotnetAvailable
tests/Audio/TranscriptionTests.cs
325lines · modecode
| 1 | using NUnit.Framework; |
| 2 | using OpenAI.Audio; |
| 3 | using OpenAI.Tests.Utility; |
| 4 | using System; |
| 5 | using System.ClientModel; |
| 6 | using System.ClientModel.Primitives; |
| 7 | using System.Collections.Generic; |
| 8 | using System.IO; |
| 9 | using System.Text; |
| 10 | using System.Threading.Tasks; |
| 11 | using System.Transactions; |
| 12 | using static OpenAI.Tests.TestHelpers; |
| 13 | |
| 14 | namespace OpenAI.Tests.Audio; |
| 15 | |
| 16 | [TestFixture(true)] |
| 17 | [TestFixture(false)] |
| 18 | [Parallelizable(ParallelScope.All)] |
| 19 | [Category("Audio")] |
| 20 | public partial class TranscriptionTests : SyncAsyncTestBase |
| 21 | { |
| 22 | public TranscriptionTests(bool isAsync) : base(isAsync) |
| 23 | { |
| 24 | } |
| 25 | |
| 26 | public enum AudioSourceKind |
| 27 | { |
| 28 | UsingStream, |
| 29 | UsingFilePath, |
| 30 | } |
| 31 | |
| 32 | [Test] |
| 33 | [TestCase(AudioSourceKind.UsingStream)] |
| 34 | [TestCase(AudioSourceKind.UsingFilePath)] |
| 35 | public async Task TranscriptionWorks(AudioSourceKind audioSourceKind) |
| 36 | { |
| 37 | AudioClient client = GetTestClient<AudioClient>(TestScenario.Audio_Whisper); |
| 38 | string filename = "audio_hello_world.mp3"; |
| 39 | string path = Path.Combine("Assets", filename); |
| 40 | AudioTranscription transcription = null; |
| 41 | |
| 42 | if (audioSourceKind == AudioSourceKind.UsingStream) |
| 43 | { |
| 44 | using FileStream inputStream = File.OpenRead(path); |
| 45 | |
| 46 | transcription = IsAsync |
| 47 | ? await client.TranscribeAudioAsync(inputStream, filename) |
| 48 | : client.TranscribeAudio(inputStream, filename); |
| 49 | } |
| 50 | else if (audioSourceKind == AudioSourceKind.UsingFilePath) |
| 51 | { |
| 52 | transcription = IsAsync |
| 53 | ? await client.TranscribeAudioAsync(path) |
| 54 | : client.TranscribeAudio(path); |
| 55 | } |
| 56 | |
| 57 | Assert.That(transcription, Is.Not.Null); |
| 58 | Assert.That(transcription.Text.ToLowerInvariant(), Contains.Substring("hello")); |
| 59 | } |
| 60 | |
| 61 | [Test] |
| 62 | [TestCase(AudioTimestampGranularities.Default)] |
| 63 | [TestCase(AudioTimestampGranularities.Word)] |
| 64 | [TestCase(AudioTimestampGranularities.Segment)] |
| 65 | [TestCase(AudioTimestampGranularities.Word | AudioTimestampGranularities.Segment)] |
| 66 | public async Task TimestampsWork(AudioTimestampGranularities granularityFlags) |
| 67 | { |
| 68 | AudioClient client = GetTestClient<AudioClient>(TestScenario.Audio_Whisper); |
| 69 | |
| 70 | using FileStream inputStream = File.OpenRead(Path.Combine("Assets", "audio_hello_world.mp3")); |
| 71 | |
| 72 | AudioTranscriptionOptions options = new() |
| 73 | { |
| 74 | ResponseFormat = AudioTranscriptionFormat.Verbose, |
| 75 | Temperature = 0.4f, |
| 76 | TimestampGranularities = granularityFlags, |
| 77 | }; |
| 78 | |
| 79 | ClientResult<AudioTranscription> transcriptionResult = IsAsync |
| 80 | ? await client.TranscribeAudioAsync(inputStream, "audio_hello_world.mp3", options) |
| 81 | : client.TranscribeAudio(inputStream, "audio_hello_world.mp3", options); |
| 82 | |
| 83 | PipelineResponse rawResponse = transcriptionResult.GetRawResponse(); |
| 84 | Assert.That(rawResponse.Content.ToString(), Is.Not.Null.And.Not.Empty); |
| 85 | |
| 86 | AudioTranscription transcription = transcriptionResult; |
| 87 | Assert.That(transcription, Is.Not.Null); |
| 88 | |
| 89 | IReadOnlyList<TranscribedWord> words = transcription.Words; |
| 90 | IReadOnlyList<TranscribedSegment> segments = transcription.Segments; |
| 91 | |
| 92 | bool wordTimestampsPresent = words?.Count > 0; |
| 93 | bool segmentTimestampsPresent = segments?.Count > 0; |
| 94 | |
| 95 | bool wordTimestampsExpected = granularityFlags.HasFlag(AudioTimestampGranularities.Word); |
| 96 | bool segmentTimestampsExpected = granularityFlags.HasFlag(AudioTimestampGranularities.Segment) |
| 97 | || granularityFlags == AudioTimestampGranularities.Default; |
| 98 | |
| 99 | Assert.That(wordTimestampsPresent, Is.EqualTo(wordTimestampsExpected)); |
| 100 | Assert.That(segmentTimestampsPresent, Is.EqualTo(segmentTimestampsExpected)); |
| 101 | |
| 102 | for (int i = 0; i < (words?.Count ?? 0); i++) |
| 103 | { |
| 104 | if (i > 0) |
| 105 | { |
| 106 | Assert.That(words[i].StartTime, Is.GreaterThanOrEqualTo(words[i - 1].EndTime)); |
| 107 | } |
| 108 | Assert.That(words[i].EndTime, Is.GreaterThan(words[i].StartTime)); |
| 109 | Assert.That(string.IsNullOrEmpty(words[i].Word), Is.False); |
| 110 | } |
| 111 | |
| 112 | for (int i = 0; i < (segments?.Count ?? 0); i++) |
| 113 | { |
| 114 | if (i > 0) |
| 115 | { |
| 116 | Assert.That(segments[i].Id, Is.GreaterThan(segments[i - 1].Id)); |
| 117 | Assert.That(segments[i].SeekOffset, Is.GreaterThan(0)); |
| 118 | Assert.That(segments[i].StartTime, Is.GreaterThanOrEqualTo(segments[i - 1].EndTime)); |
| 119 | } |
| 120 | Assert.That(segments[i].EndTime, Is.GreaterThan(segments[i].StartTime)); |
| 121 | Assert.That(string.IsNullOrEmpty(segments[i].Text), Is.False); |
| 122 | Assert.That(segments[i].TokenIds.Span.Length, Is.GreaterThan(0)); |
| 123 | foreach (int tokenId in segments[i].TokenIds.ToArray()) |
| 124 | { |
| 125 | Assert.That(tokenId, Is.GreaterThanOrEqualTo(0)); |
| 126 | } |
| 127 | Assert.That(segments[i].Temperature, Is.LessThan(-0.001f).Or.GreaterThan(0.001f)); |
| 128 | Assert.That(segments[i].AverageLogProbability, Is.LessThan(-0.001f).Or.GreaterThan(0.001f)); |
| 129 | Assert.That(segments[i].CompressionRatio, Is.LessThan(-0.001f).Or.GreaterThan(0.001f)); |
| 130 | Assert.That(segments[i].NoSpeechProbability, Is.LessThan(-0.001f).Or.GreaterThan(0.001f)); |
| 131 | } |
| 132 | } |
| 133 | |
| 134 | [Test] |
| 135 | [TestCase("text")] |
| 136 | [TestCase("json")] |
| 137 | [TestCase("verbose_json")] |
| 138 | [TestCase("srt")] |
| 139 | [TestCase("vtt")] |
| 140 | [TestCase(null)] |
| 141 | public async Task TranscriptionFormatsWork(string responseFormat) |
| 142 | { |
| 143 | AudioClient client = GetTestClient<AudioClient>(TestScenario.Audio_Whisper); |
| 144 | string path = Path.Combine("Assets", "audio_hello_world.mp3"); |
| 145 | |
| 146 | AudioTranscriptionOptions options = new() |
| 147 | { |
| 148 | ResponseFormat = responseFormat switch |
| 149 | { |
| 150 | "text" => AudioTranscriptionFormat.Text, |
| 151 | "json" => AudioTranscriptionFormat.Simple, |
| 152 | "verbose_json" => AudioTranscriptionFormat.Verbose, |
| 153 | "srt" => AudioTranscriptionFormat.Srt, |
| 154 | "vtt" => AudioTranscriptionFormat.Vtt, |
| 155 | _ => (AudioTranscriptionFormat?)null |
| 156 | } |
| 157 | }; |
| 158 | |
| 159 | AudioTranscription transcription = IsAsync |
| 160 | ? await client.TranscribeAudioAsync(path, options) |
| 161 | : client.TranscribeAudio(path, options); |
| 162 | |
| 163 | Assert.That(transcription?.Text?.ToLowerInvariant(), Does.Contain("hello")); |
| 164 | |
| 165 | |
| 166 | if (options.ResponseFormat == AudioTranscriptionFormat.Verbose) |
| 167 | { |
| 168 | Assert.That(transcription.Language, Is.EqualTo("english")); |
| 169 | Assert.That(transcription.Duration, Is.GreaterThan(TimeSpan.Zero)); |
| 170 | Assert.That(transcription.Segments, Is.Not.Empty); |
| 171 | |
| 172 | for (int i = 0; i < transcription.Segments.Count; i++) |
| 173 | { |
| 174 | TranscribedSegment segment = transcription.Segments[i]; |
| 175 | |
| 176 | if (i > 0) |
| 177 | { |
| 178 | Assert.That(segment.StartTime, Is.GreaterThanOrEqualTo(transcription.Segments[i - 1].EndTime)); |
| 179 | } |
| 180 | |
| 181 | Assert.That(segment.Id, Is.EqualTo(i)); |
| 182 | Assert.That(segment.EndTime, Is.GreaterThanOrEqualTo(segment.StartTime)); |
| 183 | Assert.That(segment.TokenIds, Is.Not.Null); |
| 184 | Assert.That(segment.TokenIds.Length, Is.GreaterThan(0)); |
| 185 | |
| 186 | Assert.That(segment.AverageLogProbability, Is.LessThan(-0.001f).Or.GreaterThan(0.001f)); |
| 187 | Assert.That(segment.CompressionRatio, Is.LessThan(-0.001f).Or.GreaterThan(0.001f)); |
| 188 | Assert.That(segment.NoSpeechProbability, Is.LessThan(-0.001f).Or.GreaterThan(0.001f)); |
| 189 | } |
| 190 | } |
| 191 | else |
| 192 | { |
| 193 | Assert.That(transcription.Duration, Is.Null); |
| 194 | Assert.That(transcription.Language, Is.Null); |
| 195 | Assert.That(transcription.Segments, Is.Not.Null.And.Empty); |
| 196 | Assert.That(transcription.Words, Is.Not.Null.And.Empty); |
| 197 | } |
| 198 | } |
| 199 | |
| 200 | [Test] |
| 201 | public async Task IncludesWork() |
| 202 | { |
| 203 | AudioClient client = GetTestClient<AudioClient>(TestScenario.Audio_Gpt_4o_Mini_Transcribe); |
| 204 | string filename = "audio_hello_world.mp3"; |
| 205 | string path = Path.Combine("Assets", filename); |
| 206 | |
| 207 | AudioTranscription transcription = await client.TranscribeAudioAsync(path, new AudioTranscriptionOptions() |
| 208 | { |
| 209 | Includes = AudioTranscriptionIncludes.Logprobs, |
| 210 | }); |
| 211 | |
| 212 | Assert.That(transcription.TranscriptionTokenLogProbabilities, Has.Count.GreaterThan(0)); |
| 213 | Assert.That(transcription.TranscriptionTokenLogProbabilities[0].Token, Is.Not.Null.And.Not.Empty); |
| 214 | Assert.That(transcription.TranscriptionTokenLogProbabilities[0].LogProbability, Is.Not.EqualTo(0)); |
| 215 | Assert.That(transcription.TranscriptionTokenLogProbabilities[0].Utf8Bytes.ToArray(), Is.Not.Null.And.Not.Empty); |
| 216 | } |
| 217 | |
| 218 | [Test] |
| 219 | public async Task StreamingIncludesWork() |
| 220 | { |
| 221 | AudioClient client = GetTestClient<AudioClient>(TestScenario.Audio_Gpt_4o_Mini_Transcribe); |
| 222 | string filename = "audio_hello_world.mp3"; |
| 223 | string path = Path.Combine("Assets", filename); |
| 224 | |
| 225 | List<AudioTokenLogProbabilityDetails> streamedDeltaLogProbs = []; |
| 226 | |
| 227 | await foreach (StreamingAudioTranscriptionUpdate update |
| 228 | in client.TranscribeAudioStreamingAsync( |
| 229 | path, |
| 230 | new AudioTranscriptionOptions() |
| 231 | { |
| 232 | Includes = AudioTranscriptionIncludes.Logprobs, |
| 233 | })) |
| 234 | { |
| 235 | if (update is StreamingAudioTranscriptionTextDeltaUpdate deltaUpdate) |
| 236 | { |
| 237 | Assert.That(deltaUpdate.TranscriptionTokenLogProbabilities, Is.Not.Null); |
| 238 | streamedDeltaLogProbs.AddRange(deltaUpdate.TranscriptionTokenLogProbabilities); |
| 239 | } |
| 240 | else if (update is StreamingAudioTranscriptionTextDoneUpdate doneUpdate) |
| 241 | { |
| 242 | Assert.That(doneUpdate.TranscriptionTokenLogProbabilities, Has.Count.GreaterThan(0)); |
| 243 | Assert.That(doneUpdate.TranscriptionTokenLogProbabilities.Count, Is.EqualTo(streamedDeltaLogProbs.Count)); |
| 244 | Assert.That(doneUpdate.TranscriptionTokenLogProbabilities[0].Token, Is.Not.Null.And.Not.Empty); |
| 245 | Assert.That(doneUpdate.TranscriptionTokenLogProbabilities[0].Token, Is.EqualTo(streamedDeltaLogProbs[0].Token)); |
| 246 | } |
| 247 | } |
| 248 | |
| 249 | Assert.That(streamedDeltaLogProbs, Has.Count.GreaterThan(0)); |
| 250 | } |
| 251 | |
| 252 | [Test] |
| 253 | public async Task BadTranscriptionRequest() |
| 254 | { |
| 255 | AudioClient client = GetTestClient<AudioClient>(TestScenario.Audio_Whisper); |
| 256 | |
| 257 | string path = Path.Combine("Assets", "audio_hello_world.mp3"); |
| 258 | |
| 259 | AudioTranscriptionOptions options = new AudioTranscriptionOptions() |
| 260 | { |
| 261 | Language = "this should cause an error" |
| 262 | }; |
| 263 | |
| 264 | Exception caughtException = null; |
| 265 | |
| 266 | try |
| 267 | { |
| 268 | _ = IsAsync |
| 269 | ? await client.TranscribeAudioAsync(path, options) |
| 270 | : client.TranscribeAudio(path, options); |
| 271 | } |
| 272 | catch (Exception ex) |
| 273 | { |
| 274 | caughtException = ex; |
| 275 | } |
| 276 | |
| 277 | Assert.That(caughtException, Is.InstanceOf<ClientResultException>()); |
| 278 | Assert.That(caughtException.Message?.ToLower(), Contains.Substring("invalid language")); |
| 279 | } |
| 280 | |
| 281 | [Test] |
| 282 | [TestCase(AudioSourceKind.UsingStream)] |
| 283 | [TestCase(AudioSourceKind.UsingFilePath)] |
| 284 | public async Task StreamingTranscriptionWorks(AudioSourceKind audioSourceKind) |
| 285 | { |
| 286 | AudioClient client = GetTestClient<AudioClient>(TestScenario.Audio_Gpt_4o_Mini_Transcribe); |
| 287 | string filename = "audio_hello_world.mp3"; |
| 288 | string path = Path.Combine("Assets", filename); |
| 289 | |
| 290 | FileStream inputStream = null; |
| 291 | |
| 292 | AsyncCollectionResult<StreamingAudioTranscriptionUpdate> streamingUpdates = null; |
| 293 | |
| 294 | if (audioSourceKind == AudioSourceKind.UsingStream) |
| 295 | { |
| 296 | inputStream = File.OpenRead(path); |
| 297 | streamingUpdates = client.TranscribeAudioStreamingAsync(inputStream, filename); |
| 298 | } |
| 299 | else if (audioSourceKind == AudioSourceKind.UsingFilePath) |
| 300 | { |
| 301 | streamingUpdates = client.TranscribeAudioStreamingAsync(path); |
| 302 | } |
| 303 | |
| 304 | StringBuilder deltaBuilder = new(); |
| 305 | |
| 306 | await foreach (StreamingAudioTranscriptionUpdate update in streamingUpdates) |
| 307 | { |
| 308 | if (update is StreamingAudioTranscriptionTextDeltaUpdate deltaUpdate) |
| 309 | { |
| 310 | deltaBuilder.Append(deltaUpdate.Delta); |
| 311 | Assert.That(deltaUpdate.TranscriptionTokenLogProbabilities, Has.Count.EqualTo(0)); |
| 312 | } |
| 313 | else if (update is StreamingAudioTranscriptionTextDoneUpdate doneUpdate) |
| 314 | { |
| 315 | Assert.That(doneUpdate.Text, Is.Not.Null.And.Not.Empty); |
| 316 | Assert.That(doneUpdate.Text, Is.EqualTo(deltaBuilder.ToString())); |
| 317 | Assert.That(doneUpdate.TranscriptionTokenLogProbabilities, Has.Count.EqualTo(0)); |
| 318 | } |
| 319 | } |
| 320 | |
| 321 | Assert.That(deltaBuilder.ToString().ToLower(), Does.Contain("hello world")); |
| 322 | |
| 323 | inputStream?.Dispose(); |
| 324 | } |
| 325 | } |
| 326 | |