openai/openai-dotnet
Publicmirrored from https://github.com/openai/openai-dotnetAvailable
examples/Realtime/Example01_AudioFromFileWithToolsAsync.cs
214lines · modeblame
b0f9e5c3Jose Arriaga Maldonado1 years ago | 1 | using NUnit.Framework; |
| 2 | using OpenAI.Images; | |
| 3 | using OpenAI.RealtimeConversation; | |
| 4 | using System; | |
| 5 | using System.ClientModel; | |
| 6 | using System.Collections.Generic; | |
| 7 | using System.IO; | |
| 8 | using System.Linq; | |
| 9 | using System.Threading.Tasks; | |
| 10 | | |
| 11 | namespace OpenAI.Examples; | |
| 12 | | |
| 13 | #pragma warning disable OPENAI002 | |
| 14 | | |
| 15 | public partial class RealtimeExamples | |
| 16 | { | |
| 17 | [Test] | |
| 18 | public async Task Example01_AudioFromFileWithToolsAsync() | |
| 19 | { | |
| 20 | RealtimeConversationClient client = new( | |
| 21 | model: "gpt-4o-realtime-preview", | |
| 22 | credential: new ApiKeyCredential(Environment.GetEnvironmentVariable("OPENAI_API_KEY"))); | |
| 23 | using RealtimeConversationSession session = await client.StartConversationSessionAsync(); | |
| 24 | | |
| 25 | // Session options control connection-wide behavior shared across all conversations, | |
| 26 | // including audio input format and voice activity detection settings. | |
| 27 | ConversationSessionOptions sessionOptions = new() | |
| 28 | { | |
| 29 | Instructions = "You are a cheerful assistant that talks like a pirate. " | |
| 30 | + "Always inform the user when you are about to call a tool. " | |
| 31 | + "Prefer to call tools whenever applicable.", | |
| 32 | Voice = ConversationVoice.Alloy, | |
| 33 | Tools = { CreateSampleWeatherTool() }, | |
| 34 | InputAudioFormat = ConversationAudioFormat.Pcm16, | |
| 35 | OutputAudioFormat = ConversationAudioFormat.Pcm16, | |
| 36 | // Input transcription options must be provided to enable transcribed feedback for input audio | |
| 37 | InputTranscriptionOptions = new() | |
| 38 | { | |
| 39 | Model = "whisper-1", | |
| 40 | }, | |
| 41 | }; | |
| 42 | | |
| 43 | await session.ConfigureSessionAsync(sessionOptions); | |
| 44 | | |
| 45 | // Conversation history or text input are provided by adding messages to the conversation. | |
| 46 | // Adding a message will not automatically begin a response turn. | |
| 47 | await session.AddItemAsync( | |
| 48 | ConversationItem.CreateUserMessage(["I'm trying to decide what to wear on my trip."])); | |
| 49 | | |
| 50 | string inputAudioPath = FindFile("Assets\\realtime_whats_the_weather_pcm16_24khz_mono.wav"); | |
| 51 | using Stream inputAudioStream = File.OpenRead(inputAudioPath); | |
| 52 | _ = session.SendInputAudioAsync(inputAudioStream); | |
| 53 | | |
| 54 | Dictionary<string, Stream> outputAudioStreamsById = []; | |
| 55 | | |
| 56 | await foreach (ConversationUpdate update in session.ReceiveUpdatesAsync()) | |
| 57 | { | |
| 58 | if (update is ConversationSessionStartedUpdate sessionStartedUpdate) | |
| 59 | { | |
| 60 | Console.WriteLine($"<<< Session started. ID: {sessionStartedUpdate.SessionId}"); | |
| 61 | Console.WriteLine(); | |
| 62 | } | |
| 63 | | |
| 64 | if (update is ConversationInputSpeechStartedUpdate speechStartedUpdate) | |
| 65 | { | |
| 66 | Console.WriteLine( | |
| 67 | $" -- Voice activity detection started at {speechStartedUpdate.AudioStartTime}"); | |
| 68 | } | |
| 69 | | |
| 70 | if (update is ConversationInputSpeechFinishedUpdate speechFinishedUpdate) | |
| 71 | { | |
| 72 | Console.WriteLine( | |
| 73 | $" -- Voice activity detection ended at {speechFinishedUpdate.AudioEndTime}"); | |
| 74 | } | |
| 75 | | |
| 76 | // Item started updates notify that the model generation process will insert a new item into | |
| 77 | // the conversation and begin streaming its content via content updates. | |
| 78 | if (update is ConversationItemStreamingStartedUpdate itemStreamingStartedUpdate) | |
| 79 | { | |
| 80 | Console.WriteLine($" -- Begin streaming of new item"); | |
| 81 | if (!string.IsNullOrEmpty(itemStreamingStartedUpdate.FunctionName)) | |
| 82 | { | |
| 83 | Console.Write($" {itemStreamingStartedUpdate.FunctionName}: "); | |
| 84 | } | |
| 85 | } | |
| 86 | | |
| 87 | if (update is ConversationItemStreamingPartDeltaUpdate deltaUpdate) | |
| 88 | { | |
| 89 | // With audio output enabled, the audio transcript of the delta update contains an approximation of | |
| 90 | // the words spoken by the model. Without audio output, the text of the delta update will contain | |
| 91 | // the segments making up the text content of a message. | |
| 92 | Console.Write(deltaUpdate.AudioTranscript); | |
| 93 | Console.Write(deltaUpdate.Text); | |
| 94 | Console.Write(deltaUpdate.FunctionArguments); | |
| 95 | if (deltaUpdate.AudioBytes is not null) | |
| 96 | { | |
| 97 | if (!outputAudioStreamsById.TryGetValue(deltaUpdate.ItemId, out Stream value)) | |
| 98 | { | |
| 99 | string filename = $"output_{sessionOptions.OutputAudioFormat}_{deltaUpdate.ItemId}.raw"; | |
| 100 | value = File.OpenWrite(filename); | |
| 101 | outputAudioStreamsById[deltaUpdate.ItemId] = value; | |
| 102 | } | |
| 103 | | |
| 104 | value.Write(deltaUpdate.AudioBytes); | |
| 105 | } | |
| 106 | } | |
| 107 | | |
| 108 | // Item finished updates arrive when all streamed data for an item has arrived and the | |
| 109 | // accumulated results are available. In the case of function calls, this is the point | |
| 110 | // where all arguments are expected to be present. | |
| 111 | if (update is ConversationItemStreamingFinishedUpdate itemStreamingFinishedUpdate) | |
| 112 | { | |
| 113 | Console.WriteLine(); | |
| 114 | Console.WriteLine($" -- Item streaming finished, item_id={itemStreamingFinishedUpdate.ItemId}"); | |
| 115 | | |
| 116 | if (itemStreamingFinishedUpdate.FunctionCallId is not null) | |
| 117 | { | |
| 118 | Console.WriteLine($" + Responding to tool invoked by item: {itemStreamingFinishedUpdate.FunctionName}"); | |
| 119 | ConversationItem functionOutputItem = ConversationItem.CreateFunctionCallOutput( | |
| 120 | callId: itemStreamingFinishedUpdate.FunctionCallId, | |
| 121 | output: "70 degrees Fahrenheit and sunny"); | |
| 122 | await session.AddItemAsync(functionOutputItem); | |
| 123 | } | |
| 124 | else if (itemStreamingFinishedUpdate.MessageContentParts?.Count > 0) | |
| 125 | { | |
| 126 | Console.Write($" + [{itemStreamingFinishedUpdate.MessageRole}]: "); | |
| 127 | foreach (ConversationContentPart contentPart in itemStreamingFinishedUpdate.MessageContentParts) | |
| 128 | { | |
| 129 | Console.Write(contentPart.AudioTranscript); | |
| 130 | } | |
| 131 | Console.WriteLine(); | |
| 132 | } | |
| 133 | } | |
| 134 | | |
| 135 | if (update is ConversationInputTranscriptionFinishedUpdate transcriptionCompletedUpdate) | |
| 136 | { | |
| 137 | Console.WriteLine(); | |
| 138 | Console.WriteLine($" -- User audio transcript: {transcriptionCompletedUpdate.Transcript}"); | |
| 139 | Console.WriteLine(); | |
| 140 | } | |
| 141 | | |
| 142 | if (update is ConversationResponseFinishedUpdate turnFinishedUpdate) | |
| 143 | { | |
| 144 | Console.WriteLine($" -- Model turn generation finished. Status: {turnFinishedUpdate.Status}"); | |
| 145 | | |
| 146 | // Here, if we processed tool calls in the course of the model turn, we finish the | |
| 147 | // client turn to resume model generation. The next model turn will reflect the tool | |
| 148 | // responses that were already provided. | |
| 149 | if (turnFinishedUpdate.CreatedItems.Any(item => item.FunctionName?.Length > 0)) | |
| 150 | { | |
| 151 | Console.WriteLine($" -- Ending client turn for pending tool responses"); | |
| 152 | await session.StartResponseAsync(); | |
| 153 | } | |
| 154 | else | |
| 155 | { | |
| 156 | break; | |
| 157 | } | |
| 158 | } | |
| 159 | | |
| 160 | if (update is ConversationErrorUpdate errorUpdate) | |
| 161 | { | |
| 162 | Console.WriteLine(); | |
| 163 | Console.WriteLine($"ERROR: {errorUpdate.Message}"); | |
| 164 | break; | |
| 165 | } | |
| 166 | } | |
| 167 | | |
| 168 | foreach ((string itemId, Stream outputAudioStream) in outputAudioStreamsById) | |
| 169 | { | |
| 170 | Console.WriteLine($"Raw audio output for {itemId}: {outputAudioStream.Length} bytes"); | |
| 171 | outputAudioStream.Dispose(); | |
| 172 | } | |
| 173 | } | |
| 174 | | |
| 175 | private static ConversationFunctionTool CreateSampleWeatherTool() | |
| 176 | { | |
86407c80Jose Arriaga Maldonado1 years ago | 177 | return new ConversationFunctionTool("get_weather_for_location") |
b0f9e5c3Jose Arriaga Maldonado1 years ago | 178 | { |
| 179 | Description = "gets the weather for a location", | |
| 180 | Parameters = BinaryData.FromString(""" | |
| 181 | { | |
| 182 | "type": "object", | |
| 183 | "properties": { | |
| 184 | "location": { | |
| 185 | "type": "string", | |
| 186 | "description": "The city and state, e.g. San Francisco, CA" | |
| 187 | }, | |
| 188 | "unit": { | |
| 189 | "type": "string", | |
| 190 | "enum": ["c","f"] | |
| 191 | } | |
| 192 | }, | |
| 193 | "required": ["location","unit"] | |
| 194 | } | |
| 195 | """) | |
| 196 | }; | |
| 197 | } | |
| 198 | | |
| 199 | private static string FindFile(string fileName) | |
| 200 | { | |
| 201 | for (string currentDirectory = Directory.GetCurrentDirectory(); | |
| 202 | currentDirectory != null && currentDirectory != Path.GetPathRoot(currentDirectory); | |
| 203 | currentDirectory = Directory.GetParent(currentDirectory)?.FullName!) | |
| 204 | { | |
| 205 | string filePath = Path.Combine(currentDirectory, fileName); | |
| 206 | if (File.Exists(filePath)) | |
| 207 | { | |
| 208 | return filePath; | |
| 209 | } | |
| 210 | } | |
| 211 | | |
| 212 | throw new FileNotFoundException($"File '{fileName}' not found."); | |
| 213 | } | |
| 214 | } |