reduction/tts.py

from __future__ import annotations

import io
import wave

import numpy as np
from piper import PiperVoice


def load_voice(voice_path: str = "models/piper/cs_CZ-jirka-medium.onnx") -> PiperVoice:
    return PiperVoice.load(voice_path)


def synthesize(voice: PiperVoice, text: str) -> tuple[np.ndarray, int]:
    """Synthesize `text` and return (mono float32 waveform in [-1,1], sample_rate)."""
    buf = io.BytesIO()
    with wave.open(buf, "wb") as wf:
        voice.synthesize_wav(text, wf)
    buf.seek(0)
    with wave.open(buf, "rb") as wf:
        sr = wf.getframerate()
        n_channels = wf.getnchannels()
        sampwidth = wf.getsampwidth()
        raw = wf.readframes(wf.getnframes())
    if sampwidth != 2:
        raise RuntimeError(f"Unexpected Piper sample width: {sampwidth} bytes")
    pcm = np.frombuffer(raw, dtype=np.int16).astype(np.float32) / 32768.0
    if n_channels > 1:
        pcm = pcm.reshape(-1, n_channels).mean(axis=1)
    return pcm, sr


def synthesize_duration(voice: PiperVoice, text: str) -> tuple[float, np.ndarray, int]:
    """Convenience: returns (duration_seconds, waveform, sample_rate)."""
    wav, sr = synthesize(voice, text)
    return len(wav) / sr, wav, sr