Files
xiaoxia-saas/apps/api/app/schemas/lipsync.py
T
xiaoxia 81180d16e5
CI/CD Pipeline / Dedup Check - skip PR tests when covered by push pipeline (push) Successful in 1s
CI/CD Pipeline / Check push changed paths (push) Successful in 12s
CI/CD Pipeline / Build Staging API Image (push) Successful in 28s
CI/CD Pipeline / Build Staging Worker Image (push) Successful in 31s
CI/CD Pipeline / Frontend Unit Tests (push) Failing after 2m9s
CI/CD Pipeline / Build Staging Web Image (push) Successful in 3m16s
CI/CD Pipeline / Integration Tests (push) Successful in 3m39s
CI/CD Pipeline / Validate - Style (push) Successful in 4m19s
CI/CD Pipeline / Deploy Staging (Watchtower auto-deploy) (push) Successful in 2m21s
CI/CD Pipeline / ACR Image Cleanup (push) Successful in 1m38s
CI/CD Pipeline / Validate - Python (mypy + alembic) (push) Successful in 8m6s
CI/CD Pipeline / Staging API Integration Tests (push) Successful in 2m51s
CI/CD Pipeline / Staging E2E Tests (push) Failing after 2m59s
CI/CD Pipeline / Unit Tests (push) Successful in 9m13s
CI/CD Pipeline / Validate - Security (push) Successful in 11m27s
CI/CD Pipeline / Production Browser E2E (push) Has been skipped
CI/CD Pipeline / Deploy Production (push) Failing after 24h3m13s
CI/CD Pipeline / Build Production Web Image (push) Failing after 24h3m17s
CI/CD Pipeline / Retag skipped Staging Web Image (push) Failing after 24h11m11s
CI/CD Pipeline / PR Build Worker Image (push) Failing after 24h14m6s
CI/CD Pipeline / PR Build API Image (push) Failing after 24h14m7s
CI/CD Pipeline / PR Build Web Image (push) Failing after 24h14m6s
CI/CD Pipeline / Retag skipped Staging API Image (push) Failing after 24h10m32s
CI/CD Pipeline / CI Gate (push) Failing after 24h2m37s
CI/CD Pipeline / Retag skipped Staging Worker Image (push) Failing after 24h10m31s
CI/CD Pipeline / Check if frontend-only change (push) Failing after 24h14m8s
CI/CD Pipeline / Canary Release to Production (push) Failing after 24h2m33s
CI/CD Pipeline / Build Production Worker Image (push) Failing after 24h2m37s
CI/CD Pipeline / Build Production API Image (push) Failing after 24h2m37s
CI/CD Pipeline / Frontend Lint (push) Failing after 24h14m2s
fix(#1898): TTS 情绪映射修正为 CosyVoice v3 官方英文枚举(P1 更正) (#1933)
fix(tts): P1 TTS emotion English enum mapping per CosyVoice v3 official spec

- EMOTION_MAP maps 7 CN labels (中立/开心/难过/生气/惊讶/恐惧/厌恶) + CN aliases + old 4 enums → 7 official English enums (neutral/happy/sad/angry/surprised/fearful/disgusted)
- instruction strictly matches official format "你说话的情感是{emotion_enum}。" (verified against 阿里云百炼 docs)
- Unknown emotion defaults to "neutral" with warning (was empty, could cause TTS failure)
- Frontend CN labels & old enums backward-compatible; case-insensitive
- Add normalize_emotion unit tests (89 cases) covering all mappings, unknown→neutral, empty→"", ASCII enum assertion in instruction
- Fix existing test assertions from Chinese descriptors to English enums
- Fixes wrong direction in PR#1932 which mapped to Chinese descriptive phrases
2026-09-15 15:08:55 +08:00

139 lines
6.1 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""对口型 API Schema 定义 — #1796 / #1809 / #1822 / #1845(配音前置).
支持三种输入模式:
1. TTS 直生模式(兼容旧版前端):传 voice_id + script_text(+ speed/emotion),
后端 Celery 异步做 TTS 合成 + MediaKit 提交。
2. 直接音频模式:传 video_url + audio_url(音频已由调用方准备好)。
3. 预合成音频模式(#1845 配音前置新主路径):前端先调 POST /lipsync/tts-preview
拿到 audio_url + sentence_timings,再在 create_job 时传 audio_url + audio_duration
+ sentence_timings,后端跳过 TTS 和时间戳计算,直接 ffprobe 校验后提交 MediaKit。
"""
from __future__ import annotations
from datetime import datetime
from typing import Optional
from pydantic import BaseModel, Field, model_validator
class LipsyncJobResponse(BaseModel):
"""对口型任务响应."""
id: str
user_id: str
project_id: str
video_url: str
audio_url: str
enable_video_loop: bool
voice_id: str = ""
script_text: str = ""
speed: float = 1.0
emotion: str = ""
mediakit_task_id: str
status: str
output_video_url: str
output_duration: float
error_message: str
error_code: str
sentence_timings: Optional[list] = None
submitted_at: Optional[datetime] = None
completed_at: Optional[datetime] = None
created_at: datetime
updated_at: datetime
class Config:
from_attributes = True
class CreateLipsyncJobRequest(BaseModel):
"""创建对口型任务请求.
三种模式(三选一):
- TTS 直生(旧版/降级):voice_id + script_text 必填;audio_url 留空。
- 直接音频:video_url + audio_url 必填。
- 预合成音频(#1845 新主路径):audio_url 必填 + 可选 audio_duration/sentence_timings;
后端同步 ffprobe 校验时长、写入 timings,直接提交 MediaKit。
"""
video_url: str = Field(..., description="人物视频 URL(MP4,≤30min,单人真人)")
# 模式 2/3:直接/预合成音频
audio_url: str = Field("", description="驱动音频 URL(mp3/aac/wav/m4a/flac);直生模式留空")
audio_duration: Optional[float] = Field(None, ge=0, description="预合成音频时长(秒),可选;后端会 ffprobe 校验")
sentence_timings: Optional[list] = Field(None, description="预合成接口返回的句子时间戳,可选;若传入则直接写入 job")
# 模式 1:TTS 直生
voice_id: str = Field("", description="音色 ID(预置音色或克隆音色 profile UUID)")
script_text: str = Field("", description="要合成的文案(直生模式必填,最长 5000 字符)")
speed: float = Field(1.0, ge=0.5, le=2.0, description="语速(0.5-2.0),默认 1.0")
emotion: str = Field(
"",
description="情绪(英文枚举 neutral/happy/sad/angry/surprised/fearful/disgusted,或中文 中立/开心/难过/生气/惊讶/恐惧/厌恶;空为默认自然)",
)
enable_video_loop: bool = Field(
True, description="音频长于视频时是否循环画面(AI数字人默认开启,防止音频长于视频被截断)"
)
project_id: str = Field("", description="项目 ID(可选)")
@model_validator(mode="after")
def _validate_input_mode(self) -> "CreateLipsyncJobRequest":
video = (self.video_url or "").strip()
if not video:
raise ValueError("video_url 不能为空")
if not video.startswith(("http://", "https://")):
raise ValueError("video_url 必须是 HTTP/HTTPS URL")
lower = video.lower().split("?")[0]
allowed_video_exts = (".mp4", ".mov", ".m4v", ".webm", ".avi", ".mkv", ".3gp")
if not any(lower.endswith(ext) for ext in allowed_video_exts):
raise ValueError("video_url 格式不支持,仅支持: " + ", ".join(allowed_video_exts))
has_audio = bool((self.audio_url or "").strip())
has_tts = bool((self.voice_id or "").strip()) and bool((self.script_text or "").strip())
if not has_audio and not has_tts:
raise ValueError(
"必须提供驱动音频:要么传 audio_url(直接/预合成音频模式),"
"要么同时传 voice_id + script_text(TTS 直生模式)"
)
if has_tts and len(self.script_text) > 5000:
raise ValueError("script_text 最长 5000 字符")
if has_audio:
au = self.audio_url.strip()
if not au.startswith(("http://", "https://")):
raise ValueError("audio_url 必须是 HTTP/HTTPS URL")
au_lower = au.lower().split("?")[0]
allowed = (".mp3", ".aac", ".wav", ".m4a", ".flac")
if not any(au_lower.endswith(ext) for ext in allowed):
raise ValueError(f"audio_url 格式不支持,仅支持: {', '.join(allowed)}")
self.audio_url = au
return self
# ── #1845 TTS 预合成接口 ────────────────────────────────────────────────
class AiAvatarTtsPreviewRequest(BaseModel):
"""步骤1「生成配音」预合成请求(同步 HTTP,~2-3s)."""
voice_id: str = Field(..., min_length=1, max_length=128, description="音色 ID")
script_text: str = Field(..., min_length=1, max_length=5000, description="要合成的文案")
speed: float = Field(1.0, ge=0.5, le=2.0, description="语速(0.5-2.0),默认 1.0")
emotion: str = Field(
"neutral",
max_length=32,
description="情绪(英文枚举 neutral/happy/sad/angry/surprised/fearful/disgusted,或中文 中立/开心/难过/生气/惊讶/恐惧/厌恶;默认 neutral)",
)
class AiAvatarTtsPreviewResponse(BaseModel):
"""TTS 预合成响应(临时 URL,24h 内有效,足够当前会话使用)."""
audio_url: str = Field(..., description="CosyVoice 临时音频 URL")
duration: float = Field(..., ge=0, description="音频总时长(秒),ffprobe 测得")
sentence_timings: list[dict] = Field(..., description="句子级精确时间戳")