luli331/oppy_google

1import io2import os3import struct4 5from google import genai6from google.genai import types7 8 9def _make_wav(pcm_data: bytes, sample_rate: int = 24000, channels: int = 1, bits_per_sample: int = 16) -> bytes:10    """Wrap raw PCM bytes in a WAV container."""11    data_size = len(pcm_data)12    byte_rate = sample_rate * channels * bits_per_sample // 813    block_align = channels * bits_per_sample // 814 15    buf = io.BytesIO()16    # RIFF header17    buf.write(b"RIFF")18    buf.write(struct.pack("<I", 36 + data_size))19    buf.write(b"WAVE")20    # fmt chunk21    buf.write(b"fmt ")22    buf.write(struct.pack("<I", 16))  # chunk size23    buf.write(struct.pack("<H", 1))   # PCM format24    buf.write(struct.pack("<H", channels))25    buf.write(struct.pack("<I", sample_rate))26    buf.write(struct.pack("<I", byte_rate))27    buf.write(struct.pack("<H", block_align))28    buf.write(struct.pack("<H", bits_per_sample))29    # data chunk30    buf.write(b"data")31    buf.write(struct.pack("<I", data_size))32    buf.write(pcm_data)33    return buf.getvalue()34 35 36def generate_speech(text: str) -> bytes:37    """Generate speech audio from text using Gemini TTS.38 39    Args:40        text: The text to convert to speech.41 42    Returns:43        WAV audio bytes ready to play in a browser.44    """45    client = genai.Client(api_key=os.getenv("GOOGLE_API_KEY"))46 47    response = client.models.generate_content(48        model="gemini-2.5-flash-preview-tts",49        contents=text,50        config=types.GenerateContentConfig(51            response_modalities=["AUDIO"],52            speech_config=types.SpeechConfig(53                voice_config=types.VoiceConfig(54                    prebuilt_voice_config=types.PrebuiltVoiceConfig(55                        voice_name="Kore",56                    )57                ),58            ),59        ),60    )61 62    audio_data = response.candidates[0].content.parts[0].inline_data.data63    return _make_wav(audio_data)64