life-echo/api/app/adapters/tts/tencent_tts.py

"""Tencent Cloud TTS adapter — implements TTSProvider port."""

import asyncio
import base64
import re
import uuid

from app.core.business_telemetry import business_span
from app.core.logging import get_logger

logger = get_logger(__name__)

# OpenAI voice name -> Tencent VoiceType ID
VOICE_MAP: dict[str, int] = {
    "alloy": 1001,
    "echo": 1002,
    "fable": 1003,
    "onyx": 1004,
    "nova": 1005,
    "shimmer": 1006,
}

# Tencent TTS API limit: ≤150 Chinese chars or ≤500 letters (英文按字母放宽到 ~480 留余量)
MAX_CHARS_PER_REQUEST_ZH = 150
MAX_CHARS_PER_REQUEST_EN = 480

# Tencent PrimaryLanguage: 1=中文（含中英混读），2=英文
PRIMARY_LANGUAGE_ZH = 1
PRIMARY_LANGUAGE_EN = 2

# Tencent ModelType: 1=新模型（覆盖大模型音色 501xxx 系列与新版精品音色）。
# 大模型音色（如 501004 月华）必须显式传 ModelType=1，否则可能被旧模型拒绝并返回空音频；
# 老精品音色（如 1001/101050 等）也接受 ModelType=1，因此无条件设置不会破坏老链路。
# 文档：https://cloud.tencent.com/document/api/1073/37995
MODEL_TYPE_LLM = 1


def _chunk_text(text: str, max_chars: int = MAX_CHARS_PER_REQUEST_ZH) -> list[str]:
    """Split text into chunks within API limit."""
    text = text.strip()
    if not text:
        return []
    if len(text) <= max_chars:
        return [text]

    chunks: list[str] = []
    # Split by sentence boundaries first
    pattern = r"[。！？.!?\n]+"
    parts = re.split(f"({pattern})", text)
    current = ""
    for i, p in enumerate(parts):
        if re.match(pattern, p):
            current += p
            if current.strip():
                chunks.append(current.strip())
                current = ""
        else:
            if len(current) + len(p) <= max_chars:
                current += p
            else:
                if current.strip():
                    chunks.append(current.strip())
                    current = ""
                # Single part exceeds limit, split by length
                while p:
                    chunk = p[:max_chars]
                    p = p[max_chars:]
                    chunks.append(chunk)
    if current.strip():
        chunks.append(current.strip())
    return chunks


class TencentTTSProvider:
    def __init__(
        self,
        secret_id: str,
        secret_key: str,
        voice_type: int = 1001,
        codec: str = "mp3",
        voice_type_en: int | None = None,
    ):
        self._secret_id = secret_id
        self._secret_key = secret_key
        self._voice_type = voice_type
        # 英文音色未单独配置时回落到 501004（月华，腾讯云大模型音色）。
        # 大模型音色 501xxx 须配合 ModelType=1（见 Tencent TextToVoice 文档）。
        self._voice_type_en = voice_type_en if voice_type_en is not None else 501004
        self._codec = codec
        self._client = None

    def _get_client(self):
        if self._client is not None:
            return self._client
        try:
            from tencentcloud.common import credential
            from tencentcloud.common.profile.client_profile import ClientProfile
            from tencentcloud.common.profile.http_profile import HttpProfile
            from tencentcloud.tts.v20190823 import tts_client

            cred = credential.Credential(self._secret_id, self._secret_key)
            http_profile = HttpProfile()
            http_profile.endpoint = "tts.tencentcloudapi.com"
            client_profile = ClientProfile()
            client_profile.httpProfile = http_profile
            self._client = tts_client.TtsClient(cred, "", client_profile)
            return self._client
        except Exception as e:
            logger.error("Tencent TTS client init failed: {}", e)
            return None

    def _synthesize_sync(
        self,
        text: str,
        voice_type: int,
        primary_language: int = PRIMARY_LANGUAGE_ZH,
    ) -> bytes:
        client = self._get_client()
        if not client:
            logger.warning(
                "tencent_tts._synthesize_sync no client provider=tencent voice_type={}",
                voice_type,
            )
            return b""
        try:
            from tencentcloud.common.exception.tencent_cloud_sdk_exception import (
                TencentCloudSDKException,
            )
            from tencentcloud.tts.v20190823 import models

            req = models.TextToVoiceRequest()
            req.Text = text
            req.SessionId = uuid.uuid4().hex
            req.VoiceType = voice_type
            req.PrimaryLanguage = primary_language
            req.SampleRate = 16000
            req.Codec = self._codec
            # 显式声明使用新模型；大模型音色（501xxx）若不带该字段会被旧模型拒绝并静默返回空音频。
            req.ModelType = MODEL_TYPE_LLM

            resp = client.TextToVoice(req)
            request_id = getattr(resp, "RequestId", None) if resp is not None else None
            audio_b64 = getattr(resp, "Audio", "") if resp is not None else ""
            if not audio_b64:
                logger.warning(
                    "tencent_tts._synthesize_sync empty audio voice_type={} "
                    "primary_language={} model_type={} request_id={}",
                    voice_type,
                    primary_language,
                    MODEL_TYPE_LLM,
                    request_id,
                )
                return b""
            audio_bytes = base64.b64decode(audio_b64)
            return audio_bytes
        except TencentCloudSDKException as e:
            logger.error(
                "Tencent TTS SDK error provider=tencent voice_type={} primary_language={} "
                "model_type={} code={} message={} request_id={} raw={}",
                voice_type,
                primary_language,
                MODEL_TYPE_LLM,
                getattr(e, "code", None),
                getattr(e, "message", None),
                getattr(e, "requestId", None),
                e,
            )
            return b""
        except Exception as e:
            logger.error(
                "Tencent TTS synthesize failed provider=tencent voice_type={} primary_language={}: {}",
                voice_type,
                primary_language,
                e,
            )
            return b""

    async def synthesize(
        self,
        text: str,
        voice: str = "alloy",
        *,
        language: str = "zh",
    ) -> bytes:
        with business_span("tts.synthesize", provider="tencent"):
            return await self._synthesize_inner(text, voice, language=language)

    async def _synthesize_inner(
        self,
        text: str,
        voice: str = "alloy",
        *,
        language: str = "zh",
    ) -> bytes:
        if not self._secret_id or not self._secret_key:
            logger.error(
                "Tencent TTS credentials not configured provider=tencent secret_id_set={} secret_key_set={}",
                bool(self._secret_id),
                bool(self._secret_key),
            )
            return b""

        # ``language`` 由 pipeline 从用户 ``language_preference`` 解析（仅 'en' / 其它→中文路径），
        # 与助手正文实际语种无关：产品规则是 TTS 主语言跟用户语言一致。
        is_en = (language or "zh").strip().lower() == "en"
        primary_language = PRIMARY_LANGUAGE_EN if is_en else PRIMARY_LANGUAGE_ZH
        default_voice = self._voice_type_en if is_en else self._voice_type
        max_chars = MAX_CHARS_PER_REQUEST_EN if is_en else MAX_CHARS_PER_REQUEST_ZH

        # Default "alloy" aligns with OpenAI TTS naming. Caller 链路里目前不会传具体音色，
        # 因此实际只走 default_voice 分支，对应 settings.tts_voice_type / tts_voice_type_en。
        v = voice.lower()
        if v == "alloy":
            voice_type = default_voice
        else:
            voice_type = VOICE_MAP.get(v, default_voice)
        chunks = _chunk_text(text, max_chars=max_chars)
        if not chunks:
            return b""

        results: list[bytes] = []
        for idx, chunk in enumerate(chunks):
            audio = await asyncio.to_thread(
                self._synthesize_sync, chunk, voice_type, primary_language
            )
            if not audio:
                logger.warning(
                    "tencent_tts.synthesize chunk failed chunk_index={} chunk_chars={} "
                    "voice_type={} primary_language={}",
                    idx,
                    len(chunk),
                    voice_type,
                    primary_language,
                )
                return b""
            logger.debug(
                "tencent_tts.synthesize chunk ok chunk_index={} chunk_chars={} audio_bytes_len={}",
                idx,
                len(chunk),
                len(audio),
            )
            results.append(audio)

        merged = b"".join(results)
        logger.debug(
            "tencent_tts.synthesize done language={} voice_type={} chunks={} total_bytes={}",
            language,
            voice_type,
            len(chunks),
            len(merged),
        )
        return merged