fix

src/Infrastructure/BotSharp.Abstraction/Realtime/IStreamChannel.cs

@@ -9,5 +9,6 @@ public interface IStreamChannel
     Task ConnectAsync(string conversationId);
     Task<StreamReceiveResult> ReceiveAsync(ArraySegment<byte> buffer, CancellationToken cancellation);
     Task SendAsync(byte[] data, CancellationToken cancellation);
+    void ClearBuffer();
     Task CloseAsync(StreamChannelStatus status, string description, CancellationToken cancellation);
 }

src/Infrastructure/BotSharp.Abstraction/Realtime/Models/RealtimeHubConnection.cs

@@ -12,6 +12,7 @@ public class RealtimeHubConnection
     public ConcurrentQueue<string> MarkQueue { get; set; } = new();
     public string CurrentAgentId { get; set; } = null!;
     public string ConversationId { get; set; } = null!;
+    public Func<string> OnModelReady { get; set; } = () => string.Empty;
     public Func<string, string> OnModelMessageReceived { get; set; } = null!;
     public Func<string> OnModelAudioResponseDone { get; set; } = null!;
     public Func<string> OnModelUserInterrupted { get; set; } = null!;

src/Infrastructure/BotSharp.Core.Realtime/Services/RealtimeHub.cs

@@ -53,8 +53,10 @@ public class RealtimeHub : IRealtimeHub
             {
                 // Not TriggerModelInference, waiting for user utter.
                 var instruction = await _completer.UpdateSession(_conn);
-
+                var data = _conn.OnModelReady();
+                await responseToUser(data);
                 await HookEmitter.Emit<IRealtimeHook>(_services, async hook => await hook.OnModeReady(agent, _completer));
+
             },
             onModelAudioDeltaReceived: async (audioDeltaData, itemId) =>
             {

src/Infrastructure/BotSharp.Core.Realtime/Services/WaveStremChannel.cs

@@ -1,7 +1,6 @@
+using System.Collections.Concurrent;
 using BotSharp.Abstraction.Realtime.Enums;
 using NAudio.Wave;
-using System.Collections.Concurrent;
-using System.IO;
 
 namespace BotSharp.Core.Realtime.Services;
 
@@ -11,7 +10,7 @@ public class WaveStremChannel : IStreamChannel
     private WaveInEvent _waveIn;
     private WaveOutEvent _waveOut;
     private BufferedWaveProvider _bufferedWaveProvider;
-    private readonly ConcurrentQueue<byte[]> _audioBufferQueue = new ConcurrentQueue<byte[]>();
+    private readonly ConcurrentQueue<byte[]> _audioBufferQueue = [];
     private readonly ILogger _logger;
 
     public WaveStremChannel(IServiceProvider services, ILogger<WaveStremChannel> logger)
@@ -39,7 +38,7 @@ public class WaveStremChannel : IStreamChannel
         // Initialize audio output for streaming
         var waveFormat = new WaveFormat(24000, 16, 1); // 24000 Hz, 16-bit PCM, Mono
         _bufferedWaveProvider = new BufferedWaveProvider(waveFormat);
-        _bufferedWaveProvider.BufferLength = 1024 * 512; // Buffer length
+        _bufferedWaveProvider.BufferLength = 1024 * 1024; // Buffer length
         _bufferedWaveProvider.DiscardOnBufferOverflow = true;
         
         _waveOut = new WaveOutEvent();
@@ -83,6 +82,11 @@ public class WaveStremChannel : IStreamChannel
         return Task.CompletedTask;
     }
 
+    public void ClearBuffer()
+    {
+        _bufferedWaveProvider?.ClearBuffer();
+    }
+
     private void WaveIn_DataAvailable(object? sender, WaveInEventArgs e)
     {
         // Add the buffer to the queue

src/Plugins/BotSharp.Plugin.OpenAI/Providers/Realtime/RealTimeCompletionProvider.cs

@@ -15,6 +15,7 @@ public class RealTimeCompletionProvider : IRealTimeCompletion
     protected readonly OpenAiSettings _settings;
     protected readonly IServiceProvider _services;
     protected readonly ILogger<RealTimeCompletionProvider> _logger;
+    private readonly BotSharpOptions _options;
 
     protected string _model = "gpt-4o-mini-realtime-preview";
     private ClientWebSocket _webSocket;
@@ -22,11 +23,13 @@ public class RealTimeCompletionProvider : IRealTimeCompletion
     public RealTimeCompletionProvider(
         OpenAiSettings settings,
         ILogger<RealTimeCompletionProvider> logger,
-        IServiceProvider services)
+        IServiceProvider services,
+        BotSharpOptions options)
     {
         _settings = settings;
         _logger = logger;
         _services = services;
+        _options = options;
     }
 
     public async Task Connect(RealtimeHubConnection conn,
@@ -45,6 +48,7 @@ public class RealTimeCompletionProvider : IRealTimeCompletion
         var settingsService = _services.GetRequiredService<ILlmProviderService>();
         var settings = settingsService.GetSetting(Provider, _model);
 
+        _webSocket?.Dispose();
         _webSocket = new ClientWebSocket();
         _webSocket.Options.SetRequestHeader("Authorization", $"Bearer {settings.ApiKey}");
         _webSocket.Options.SetRequestHeader("OpenAI-Beta", "realtime=v1");
@@ -141,7 +145,7 @@ public class RealTimeCompletionProvider : IRealTimeCompletion
         Action<RoleDialogModel> onUserAudioTranscriptionCompleted,
         Action onUserInterrupted)
     {
-        var buffer = new byte[1024 * 32];
+        var buffer = new byte[1024 * 1024 * 32];
         // Model response timeout
         var timeout = 30;
         WebSocketReceiveResult? result = default;
@@ -276,7 +280,7 @@ public class RealTimeCompletionProvider : IRealTimeCompletion
 
         if (message is not string data)
         {
-            data = JsonSerializer.Serialize(message, BotSharpOptions.defaultJsonOptions);
+            data = JsonSerializer.Serialize(message, _options.JsonSerializerOptions);
         }
 
         var buffer = Encoding.UTF8.GetBytes(data);
@@ -291,12 +295,9 @@ public class RealTimeCompletionProvider : IRealTimeCompletion
 
         var agentService = _services.GetRequiredService<IAgentService>();
         var agent = await agentService.LoadAgent(conn.CurrentAgentId);
-
-        var client = ProviderHelper.GetClient(Provider, _model, _services);
-        var chatClient = client.GetChatClient(_model);
         var (prompt, messages, options) = PrepareOptions(agent, []);
 
-        var instruction = messages.FirstOrDefault()?.Content.FirstOrDefault()?.Text ?? agent.Description;
+        var instruction = messages.FirstOrDefault()?.Content.FirstOrDefault()?.Text ?? agent?.Description;
         var functions = options.Tools.Select(x =>
         {
             var fn = new FunctionDef

tests/BotSharp.Test.RealtimeVoice/Program.cs

@@ -3,7 +3,6 @@ using BotSharp.Abstraction.Conversations.Models;
 using BotSharp.Abstraction.Conversations;
 using BotSharp.OpenAPI;
 using System.Text.Json;
-using Google.Ai.Generativelanguage.V1Beta2;
 
 var services = ServiceBuilder.CreateHostBuilder();
 var channel = services.GetRequiredService<IStreamChannel>();
@@ -27,18 +26,11 @@ var hub = services.GetRequiredService<IRealtimeHub>();
 var conn = hub.SetHubConnection(conv.Id);
 var completer = hub.SetCompleter("openai");
 
-await hub.ConnectToModel(async data =>
-{
-    var response = JsonSerializer.Deserialize<ModelResponseEvent>(data);
-    if (response.Event == "media")
+conn.OnModelReady = () =>
+    JsonSerializer.Serialize(new
     {
-        var message = JsonSerializer.Deserialize<ModelResponseMediaEvent>(data);
-        await channel.SendAsync(Convert.FromBase64String(message.Media), CancellationToken.None);
-    }
-});
-
-StreamReceiveResult result;
-var buffer = new byte[1024 * 8];
+        @event = "init"
+    });
 
 conn.OnModelMessageReceived = message =>
     JsonSerializer.Serialize(new
@@ -60,6 +52,23 @@ conn.OnModelUserInterrupted = () =>
         @event = "clear"
     });
 
+await hub.ConnectToModel(async data =>
+{
+    var response = JsonSerializer.Deserialize<ModelResponseEvent>(data);
+    if (response.Event == "clear")
+    {
+        channel.ClearBuffer();
+    }
+    else if (response.Event == "media")
+    {
+        var message = JsonSerializer.Deserialize<ModelResponseMediaEvent>(data);
+        await channel.SendAsync(Convert.FromBase64String(message.Media), CancellationToken.None);
+    }
+});
+
+StreamReceiveResult result;
+var buffer = new byte[1024 * 8];
+
 do
 {
     var seg = new ArraySegment<byte>(buffer);
@@ -75,22 +84,38 @@ do
 int CalculateAudioLevel(byte[] buffer, int bytesRecorded)
 {
     // Simple audio level calculation (RMS)
-    int sum = 0;
+    int bytesPerSample = 2; // 16-bit PCM = 2 bytes per sample
+    int sampleCount = bytesRecorded / bytesPerSample;
+    if (sampleCount == 0) return 0;
+
+    double sum = 0;
     for (int i = 0; i < bytesRecorded; i += 2)
     {
         if (i + 1 < bytesRecorded)
         {
             short sample = (short)((buffer[i + 1] << 8) | buffer[i]);
-            sum += Math.Abs(sample);
+            double normalized = sample / (short.MaxValue * 1.0 + 1);
+            sum += normalized * normalized;
         }
     }
-    return bytesRecorded > 0 ? sum / (bytesRecorded / 2) : 0;
+
+    double rms = Math.Sqrt(sum / sampleCount);
+    double db = 20 * Math.Log10(rms);
+
+    if (double.IsInfinity(db) || double.IsNaN(db))
+    {
+        return 0;
+    }
+
+    db = Math.Clamp(db, -100, 0);
+    return (int)((db + 100) * 1);
 }
 
 void DisplayAudioLevel(int level)
 {
+    const int sep = 50;
     // Normalize level to 0-50 range for display
-    int displayLevel = Math.Min(50, level / 100);
+    int displayLevel = (level * sep) / 100;
 
     // Clear the current line
     Console.Write("\r" + new string(' ', 60));
@@ -98,6 +123,6 @@ void DisplayAudioLevel(int level)
     // Display audio level as a bar
     Console.Write("\rMicrophone: [");
     Console.Write(new string('#', displayLevel));
-    Console.Write(new string(' ', 50 - displayLevel));
+    Console.Write(new string(' ', sep - displayLevel));
     Console.Write("]");
 }