Update app.py

app.py

@@ -42,9 +42,9 @@ def custom_split_text(text: str) -> list:
     Custom splitting:
       - Start with a chunk size of 2 words.
       - For each chunk, if a period (".") is found in any word (except if it’s the very last word),
-        then split the chunk at that word (i.e. include words up to and including that word).
+        then split the chunk at that word (include words up to that word).
       - Otherwise, use the current chunk size.
-      - For subsequent chunks, increase the chunk size by 2 (i.e. 2, 4, 6, …).
+      - For subsequent chunks, increase the chunk size by 2.
       - If there are fewer than the desired number of words for a full chunk, add all remaining words.
     """
     words = text.split()
@@ -56,26 +56,24 @@ def custom_split_text(text: str) -> list:
         if candidate_end > len(words):
             candidate_end = len(words)
         chunk_words = words[start:candidate_end]
-        # Look for a period in the chunk (from right to left)
+        # Look for a period in any word except the last one.
         split_index = None
-        for i in reversed(range(len(chunk_words))):
+        for i in range(len(chunk_words) - 1):
             if '.' in chunk_words[i]:
                 split_index = i
                 break
-        if split_index is not None and split_index != len(chunk_words) - 1:
-            # If a period is found and it’s not the last word in the chunk,
-            # adjust the chunk so it ends at that word.
+        if split_index is not None:
             candidate_end = start + split_index + 1
             chunk_words = words[start:candidate_end]
         chunks.append(" ".join(chunk_words))
         start = candidate_end
-        chunk_size += 2  # Increase by 2 (added, not multiplied)
+        chunk_size += 2  # Increase the chunk size by 2 for the next iteration.
     return chunks
 
 
 def audio_tensor_to_pcm_bytes(audio_tensor: torch.Tensor) -> bytes:
     """
-    Convert a torch.FloatTensor (with values assumed in [-1, 1]) to raw 16-bit PCM bytes.
+    Convert a torch.FloatTensor (with values in [-1, 1]) to raw 16-bit PCM bytes.
     """
     audio_np = audio_tensor.cpu().numpy()
     if audio_np.ndim > 1:
@@ -93,7 +91,7 @@ def tts_streaming(text: str, voice: str = "af_heart", speed: float = 1.0):
     """
     Streaming TTS endpoint that returns a continuous WAV stream.
     
-    The endpoint first yields a WAV header (with a dummy length) then yields raw PCM data
+    This endpoint first yields a WAV header (with a dummy data length) and then yields raw PCM data
     for each text chunk as soon as it is generated.
     """
     chunks = custom_split_text(text)
@@ -118,6 +116,7 @@ def tts_streaming(text: str, voice: str = "af_heart", speed: float = 1.0):
                         print(f"Chunk {i}: No audio generated")
             except Exception as e:
                 print(f"Error processing chunk {i}: {e}")
+
     return StreamingResponse(
         audio_generator(),
         media_type="audio/wav",
@@ -147,7 +146,7 @@ def tts_full(text: str, voice: str = "af_heart", speed: float = 1.0):
 
     sample_rate = 24000
     num_channels = 1
-    sample_width = 2  # 16-bit PCM
+    sample_width = 2  # 16-bit PCM -> 2 bytes per sample
     wav_io = io.BytesIO()
     with wave.open(wav_io, "wb") as wav_file:
         wav_file.setnchannels(num_channels)
@@ -163,14 +162,13 @@ def tts_full(text: str, voice: str = "af_heart", speed: float = 1.0):
 def index():
     """
     HTML demo page for Kokoro TTS.
-    
+
     Two playback methods are provided:
-      - "Play Full TTS" uses a standard <audio> element.
-      - "Play Streaming TTS" uses the Web Audio API (via a ScriptProcessorNode) to stream
-         the raw PCM data as it arrives. This method first reads the WAV header (44 bytes)
-         then continuously pulls in PCM data, converts it to Float32, and plays it.
+      - "Play Streaming TTS" sets the <audio> element's src to the streaming endpoint.
+      - "Play Full TTS" sets the <audio> element's src to the full synthesis endpoint.
+    The browser’s native playback handles streaming (progressive download) of the WAV data.
     """
-    return r"""
+    return """
 <!DOCTYPE html>
 <html>
 <head>
@@ -183,109 +181,29 @@ def index():
     <input type="text" id="voice" value="af_heart"><br>
     <label for="speed">Speed:</label>
     <input type="number" step="0.1" id="speed" value="1.0"><br><br>
-    <button onclick="startStreaming()">Play Streaming TTS (Web Audio API)</button>
-    <button onclick="playFull()">Play Full TTS (Standard Audio)</button>
+    <button onclick="playStreaming()">Play Streaming TTS</button>
+    <button onclick="playFull()">Play Full TTS</button>
     <br><br>
-    <audio id="fullAudio" controls></audio>
+    <audio id="audioPlayer" controls autoplay></audio>
     <script>
-    // Function to play full TTS by simply setting the <audio> element's source.
-    function playFull() {
-        const text = document.getElementById('text').value;
-        const voice = document.getElementById('voice').value;
-        const speed = document.getElementById('speed').value;
-        const audio = document.getElementById('fullAudio');
-        audio.src = `/tts/full?text=${encodeURIComponent(text)}&voice=${encodeURIComponent(voice)}&speed=${speed}`;
-        audio.play();
-    }
-
-    // Function to stream audio using the Web Audio API.
-    async function startStreaming() {
-        const text = document.getElementById('text').value;
-        const voice = document.getElementById('voice').value;
-        const speed = document.getElementById('speed').value;
-        const response = await fetch(`/tts/streaming?text=${encodeURIComponent(text)}&voice=${encodeURIComponent(voice)}&speed=${speed}`);
-        if (!response.body) {
-            alert("Streaming not supported in this browser.");
-            return;
-        }
-        
-        const reader = response.body.getReader();
-        const audioContext = new (window.AudioContext || window.webkitAudioContext)();
-        // Create a ScriptProcessorNode (buffer size of 4096 samples)
-        const scriptNode = audioContext.createScriptProcessor(4096, 1, 1);
-        let bufferQueue = [];
-        let currentBuffer = new Float32Array(0);
-        let headerRead = false;
-        let headerBytes = new Uint8Array(0);
-
-        // Helper: Convert Int16 PCM (little-endian) to Float32.
-        function int16ToFloat32(buffer) {
-            const len = buffer.length;
-            const floatBuffer = new Float32Array(len);
-            for (let i = 0; i < len; i++) {
-                floatBuffer[i] = buffer[i] / 32767;
-            }
-            return floatBuffer;
+        function playStreaming() {
+            const text = document.getElementById('text').value;
+            const voice = document.getElementById('voice').value;
+            const speed = document.getElementById('speed').value;
+            const audio = document.getElementById('audioPlayer');
+            // Simply point the audio element to the streaming endpoint.
+            audio.src = `/tts/streaming?text=${encodeURIComponent(text)}&voice=${encodeURIComponent(voice)}&speed=${speed}`;
+            audio.play();
         }
-        
-        scriptNode.onaudioprocess = function(e) {
-            const output = e.outputBuffer.getChannelData(0);
-            let offset = 0;
-            while (offset < output.length) {
-                if (currentBuffer.length === 0) {
-                    if (bufferQueue.length > 0) {
-                        currentBuffer = bufferQueue.shift();
-                    } else {
-                        // If no data is available, output silence.
-                        for (let i = offset; i < output.length; i++) {
-                            output[i] = 0;
-                        }
-                        break;
-                    }
-                }
-                const needed = output.length - offset;
-                const available = currentBuffer.length;
-                const toCopy = Math.min(needed, available);
-                output.set(currentBuffer.slice(0, toCopy), offset);
-                offset += toCopy;
-                if (toCopy < currentBuffer.length) {
-                    currentBuffer = currentBuffer.slice(toCopy);
-                } else {
-                    currentBuffer = new Float32Array(0);
-                }
-            }
-        };
-        scriptNode.connect(audioContext.destination);
-
-        // Read the response stream.
-        while (true) {
-            const { done, value } = await reader.read();
-            if (done) break;
-            let chunk = value;
-            // First, accumulate the 44-byte WAV header.
-            if (!headerRead) {
-                let combined = new Uint8Array(headerBytes.length + chunk.length);
-                combined.set(headerBytes);
-                combined.set(chunk, headerBytes.length);
-                if (combined.length >= 44) {
-                    headerBytes = combined.slice(0, 44);
-                    headerRead = true;
-                    // Remove the header bytes from the chunk.
-                    chunk = combined.slice(44);
-                } else {
-                    headerBytes = combined;
-                    continue;
-                }
-            }
-            // Make sure the chunk length is even (2 bytes per sample).
-            if (chunk.length % 2 !== 0) {
-                chunk = chunk.slice(0, chunk.length - 1);
-            }
-            const int16Buffer = new Int16Array(chunk.buffer, chunk.byteOffset, chunk.byteLength / 2);
-            const floatBuffer = int16ToFloat32(int16Buffer);
-            bufferQueue.push(floatBuffer);
+        function playFull() {
+            const text = document.getElementById('text').value;
+            const voice = document.getElementById('voice').value;
+            const speed = document.getElementById('speed').value;
+            const audio = document.getElementById('audioPlayer');
+            // Simply point the audio element to the full synthesis endpoint.
+            audio.src = `/tts/full?text=${encodeURIComponent(text)}&voice=${encodeURIComponent(voice)}&speed=${speed}`;
+            audio.play();
         }
-    }
     </script>
 </body>
 </html>