| 1 | """Alibaba CosyVoice backend for narration audio generation.""" |
| 2 | |
| 3 | from __future__ import annotations |
| 4 | |
| 5 | import os |
| 6 | from pathlib import Path |
| 7 | |
| 8 | from tts_backends.backend_common import download_audio, extension_from_format, post_json, read_api_key |
| 9 | |
| 10 | |
| 11 | DEFAULT_ENDPOINT = "https://dashscope.aliyuncs.com/api/v1/services/audio/tts/SpeechSynthesizer" |
| 12 | DEFAULT_MODEL = "cosyvoice-v3-flash" |
| 13 | |
| 14 | |
| 15 | def output_extension(audio_format: str) -> str: |
| 16 | return extension_from_format(audio_format) |
| 17 | |
| 18 | |
| 19 | def read_cosyvoice_api_key(env_name: str) -> str: |
| 20 | env_names = tuple(dict.fromkeys([env_name, "COSYVOICE_API_KEY", "DASHSCOPE_API_KEY"])) |
| 21 | return read_api_key(*env_names, label="CosyVoice/DashScope") |
| 22 | |
| 23 | |
| 24 | def resolve_url(base_url: str | None = None) -> str: |
| 25 | base = (base_url or os.environ.get("COSYVOICE_TTS_BASE_URL") or DEFAULT_ENDPOINT).rstrip("/") |
| 26 | if base.endswith("/SpeechSynthesizer"): |
| 27 | return base |
| 28 | return base + "/api/v1/services/audio/tts/SpeechSynthesizer" |
| 29 | |
| 30 | |
| 31 | def generate( |
| 32 | text: str, |
| 33 | output_path: Path, |
| 34 | *, |
| 35 | api_key: str, |
| 36 | voice_id: str, |
| 37 | model: str, |
| 38 | audio_format: str, |
| 39 | sample_rate: int, |
| 40 | volume: int | None, |
| 41 | rate: float | None, |
| 42 | pitch: float | None, |
| 43 | instruction: str | None, |
| 44 | language_hint: str | None, |
| 45 | base_url: str | None, |
| 46 | ) -> None: |
| 47 | input_payload: dict[str, object] = { |
| 48 | "text": text, |
| 49 | "voice": voice_id, |
| 50 | "format": audio_format, |
| 51 | "sample_rate": sample_rate, |
| 52 | } |
| 53 | if volume is not None: |
| 54 | input_payload["volume"] = volume |
| 55 | if rate is not None: |
| 56 | input_payload["rate"] = rate |
| 57 | if pitch is not None: |
| 58 | input_payload["pitch"] = pitch |
| 59 | if instruction: |
| 60 | input_payload["instruction"] = instruction |
| 61 | if language_hint: |
| 62 | input_payload["language_hints"] = [language_hint] |
| 63 | |
| 64 | data = post_json( |
| 65 | resolve_url(base_url), |
| 66 | headers={"Authorization": f"Bearer {api_key}"}, |
| 67 | payload={ |
| 68 | "model": model, |
| 69 | "input": input_payload, |
| 70 | }, |
| 71 | timeout=180, |
| 72 | ) |
| 73 | audio = (data.get("output") or {}).get("audio") or {} |
| 74 | audio_url = audio.get("url") |
| 75 | if not audio_url: |
| 76 | raise RuntimeError(f"CosyVoice response missing audio URL: {data}") |
| 77 | download_audio(audio_url, output_path) |
| 78 | |
| 79 | |
| 80 | def print_voices() -> None: |
| 81 | print("CosyVoice voices are selected by voice.") |
| 82 | print("Use a system voice name or a cloned/designed voice_id from CosyVoice.") |
| 83 | print("Example system voice from Alibaba Cloud docs: longanyang") |
| 84 |