Upload chatbot/tts.py with huggingface_hub

chatbot/tts.py

@@ -0,0 +1,36 @@
+"""Text-to-speech via Kokoro ONNX (82M parameter model, runs on CPU)."""
+
+import io
+import numpy as np
+
+_kokoro_instance = None
+
+
+def _get_kokoro():
+    global _kokoro_instance
+    if _kokoro_instance is not None:
+        return _kokoro_instance
+
+    from huggingface_hub import hf_hub_download
+    from kokoro_onnx import Kokoro
+
+    model_path = hf_hub_download(
+        repo_id="fastrtc/kokoro-onnx", filename="kokoro-v1.0.onnx"
+    )
+    voices_path = hf_hub_download(
+        repo_id="fastrtc/kokoro-onnx", filename="voices-v1.0.bin"
+    )
+    _kokoro_instance = Kokoro(model_path, voices_path)
+    return _kokoro_instance
+
+
+def synthesize(text: str, voice: str = "af_sarah", speed: float = 1.0) -> bytes:
+    """Convert text to WAV audio bytes."""
+    kokoro = _get_kokoro()
+    samples, sample_rate = kokoro.create(text, voice=voice, speed=speed, lang="en-us")
+
+    buf = io.BytesIO()
+    import soundfile as sf
+    sf.write(buf, samples, sample_rate, format="WAV")
+    buf.seek(0)
+    return buf.read()