main.py 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517
  1. import hashlib
  2. import json
  3. import logging
  4. import os
  5. import uuid
  6. from functools import lru_cache
  7. from pathlib import Path
  8. import requests
  9. from open_webui.config import (
  10. AUDIO_STT_ENGINE,
  11. AUDIO_STT_MODEL,
  12. AUDIO_STT_OPENAI_API_BASE_URL,
  13. AUDIO_STT_OPENAI_API_KEY,
  14. AUDIO_TTS_API_KEY,
  15. AUDIO_TTS_ENGINE,
  16. AUDIO_TTS_MODEL,
  17. AUDIO_TTS_OPENAI_API_BASE_URL,
  18. AUDIO_TTS_OPENAI_API_KEY,
  19. AUDIO_TTS_SPLIT_ON,
  20. AUDIO_TTS_VOICE,
  21. CACHE_DIR,
  22. CORS_ALLOW_ORIGIN,
  23. DEVICE_TYPE,
  24. WHISPER_MODEL,
  25. WHISPER_MODEL_AUTO_UPDATE,
  26. WHISPER_MODEL_DIR,
  27. AppConfig,
  28. )
  29. from open_webui.constants import ERROR_MESSAGES
  30. from open_webui.env import SRC_LOG_LEVELS
  31. from fastapi import Depends, FastAPI, File, HTTPException, Request, UploadFile, status
  32. from fastapi.middleware.cors import CORSMiddleware
  33. from fastapi.responses import FileResponse
  34. from pydantic import BaseModel
  35. from open_webui.utils.utils import get_admin_user, get_current_user, get_verified_user
  36. log = logging.getLogger(__name__)
  37. log.setLevel(SRC_LOG_LEVELS["AUDIO"])
  38. app = FastAPI()
  39. app.add_middleware(
  40. CORSMiddleware,
  41. allow_origins=CORS_ALLOW_ORIGIN,
  42. allow_credentials=True,
  43. allow_methods=["*"],
  44. allow_headers=["*"],
  45. )
  46. app.state.config = AppConfig()
  47. app.state.config.STT_OPENAI_API_BASE_URL = AUDIO_STT_OPENAI_API_BASE_URL
  48. app.state.config.STT_OPENAI_API_KEY = AUDIO_STT_OPENAI_API_KEY
  49. app.state.config.STT_ENGINE = AUDIO_STT_ENGINE
  50. app.state.config.STT_MODEL = AUDIO_STT_MODEL
  51. app.state.config.TTS_OPENAI_API_BASE_URL = AUDIO_TTS_OPENAI_API_BASE_URL
  52. app.state.config.TTS_OPENAI_API_KEY = AUDIO_TTS_OPENAI_API_KEY
  53. app.state.config.TTS_ENGINE = AUDIO_TTS_ENGINE
  54. app.state.config.TTS_MODEL = AUDIO_TTS_MODEL
  55. app.state.config.TTS_VOICE = AUDIO_TTS_VOICE
  56. app.state.config.TTS_API_KEY = AUDIO_TTS_API_KEY
  57. app.state.config.TTS_SPLIT_ON = AUDIO_TTS_SPLIT_ON
  58. # setting device type for whisper model
  59. whisper_device_type = DEVICE_TYPE if DEVICE_TYPE and DEVICE_TYPE == "cuda" else "cpu"
  60. log.info(f"whisper_device_type: {whisper_device_type}")
  61. SPEECH_CACHE_DIR = Path(CACHE_DIR).joinpath("./audio/speech/")
  62. SPEECH_CACHE_DIR.mkdir(parents=True, exist_ok=True)
  63. class TTSConfigForm(BaseModel):
  64. OPENAI_API_BASE_URL: str
  65. OPENAI_API_KEY: str
  66. API_KEY: str
  67. ENGINE: str
  68. MODEL: str
  69. VOICE: str
  70. SPLIT_ON: str
  71. class STTConfigForm(BaseModel):
  72. OPENAI_API_BASE_URL: str
  73. OPENAI_API_KEY: str
  74. ENGINE: str
  75. MODEL: str
  76. class AudioConfigUpdateForm(BaseModel):
  77. tts: TTSConfigForm
  78. stt: STTConfigForm
  79. from pydub import AudioSegment
  80. from pydub.utils import mediainfo
  81. def is_mp4_audio(file_path):
  82. """Check if the given file is an MP4 audio file."""
  83. if not os.path.isfile(file_path):
  84. print(f"File not found: {file_path}")
  85. return False
  86. info = mediainfo(file_path)
  87. if (
  88. info.get("codec_name") == "aac"
  89. and info.get("codec_type") == "audio"
  90. and info.get("codec_tag_string") == "mp4a"
  91. ):
  92. return True
  93. return False
  94. def convert_mp4_to_wav(file_path, output_path):
  95. """Convert MP4 audio file to WAV format."""
  96. audio = AudioSegment.from_file(file_path, format="mp4")
  97. audio.export(output_path, format="wav")
  98. print(f"Converted {file_path} to {output_path}")
  99. @app.get("/config")
  100. async def get_audio_config(user=Depends(get_admin_user)):
  101. return {
  102. "tts": {
  103. "OPENAI_API_BASE_URL": app.state.config.TTS_OPENAI_API_BASE_URL,
  104. "OPENAI_API_KEY": app.state.config.TTS_OPENAI_API_KEY,
  105. "API_KEY": app.state.config.TTS_API_KEY,
  106. "ENGINE": app.state.config.TTS_ENGINE,
  107. "MODEL": app.state.config.TTS_MODEL,
  108. "VOICE": app.state.config.TTS_VOICE,
  109. "SPLIT_ON": app.state.config.TTS_SPLIT_ON,
  110. },
  111. "stt": {
  112. "OPENAI_API_BASE_URL": app.state.config.STT_OPENAI_API_BASE_URL,
  113. "OPENAI_API_KEY": app.state.config.STT_OPENAI_API_KEY,
  114. "ENGINE": app.state.config.STT_ENGINE,
  115. "MODEL": app.state.config.STT_MODEL,
  116. },
  117. }
  118. @app.post("/config/update")
  119. async def update_audio_config(
  120. form_data: AudioConfigUpdateForm, user=Depends(get_admin_user)
  121. ):
  122. app.state.config.TTS_OPENAI_API_BASE_URL = form_data.tts.OPENAI_API_BASE_URL
  123. app.state.config.TTS_OPENAI_API_KEY = form_data.tts.OPENAI_API_KEY
  124. app.state.config.TTS_API_KEY = form_data.tts.API_KEY
  125. app.state.config.TTS_ENGINE = form_data.tts.ENGINE
  126. app.state.config.TTS_MODEL = form_data.tts.MODEL
  127. app.state.config.TTS_VOICE = form_data.tts.VOICE
  128. app.state.config.TTS_SPLIT_ON = form_data.tts.SPLIT_ON
  129. app.state.config.STT_OPENAI_API_BASE_URL = form_data.stt.OPENAI_API_BASE_URL
  130. app.state.config.STT_OPENAI_API_KEY = form_data.stt.OPENAI_API_KEY
  131. app.state.config.STT_ENGINE = form_data.stt.ENGINE
  132. app.state.config.STT_MODEL = form_data.stt.MODEL
  133. return {
  134. "tts": {
  135. "OPENAI_API_BASE_URL": app.state.config.TTS_OPENAI_API_BASE_URL,
  136. "OPENAI_API_KEY": app.state.config.TTS_OPENAI_API_KEY,
  137. "API_KEY": app.state.config.TTS_API_KEY,
  138. "ENGINE": app.state.config.TTS_ENGINE,
  139. "MODEL": app.state.config.TTS_MODEL,
  140. "VOICE": app.state.config.TTS_VOICE,
  141. "SPLIT_ON": app.state.config.TTS_SPLIT_ON,
  142. },
  143. "stt": {
  144. "OPENAI_API_BASE_URL": app.state.config.STT_OPENAI_API_BASE_URL,
  145. "OPENAI_API_KEY": app.state.config.STT_OPENAI_API_KEY,
  146. "ENGINE": app.state.config.STT_ENGINE,
  147. "MODEL": app.state.config.STT_MODEL,
  148. },
  149. }
  150. @app.post("/speech")
  151. async def speech(request: Request, user=Depends(get_verified_user)):
  152. body = await request.body()
  153. name = hashlib.sha256(body).hexdigest()
  154. file_path = SPEECH_CACHE_DIR.joinpath(f"{name}.mp3")
  155. file_body_path = SPEECH_CACHE_DIR.joinpath(f"{name}.json")
  156. # Check if the file already exists in the cache
  157. if file_path.is_file():
  158. return FileResponse(file_path)
  159. if app.state.config.TTS_ENGINE == "openai":
  160. headers = {}
  161. headers["Authorization"] = f"Bearer {app.state.config.TTS_OPENAI_API_KEY}"
  162. headers["Content-Type"] = "application/json"
  163. try:
  164. body = body.decode("utf-8")
  165. body = json.loads(body)
  166. body["model"] = app.state.config.TTS_MODEL
  167. body = json.dumps(body).encode("utf-8")
  168. except Exception:
  169. pass
  170. r = None
  171. try:
  172. r = requests.post(
  173. url=f"{app.state.config.TTS_OPENAI_API_BASE_URL}/audio/speech",
  174. data=body,
  175. headers=headers,
  176. stream=True,
  177. )
  178. r.raise_for_status()
  179. # Save the streaming content to a file
  180. with open(file_path, "wb") as f:
  181. for chunk in r.iter_content(chunk_size=8192):
  182. f.write(chunk)
  183. with open(file_body_path, "w") as f:
  184. json.dump(json.loads(body.decode("utf-8")), f)
  185. # Return the saved file
  186. return FileResponse(file_path)
  187. except Exception as e:
  188. log.exception(e)
  189. error_detail = "Open WebUI: Server Connection Error"
  190. if r is not None:
  191. try:
  192. res = r.json()
  193. if "error" in res:
  194. error_detail = f"External: {res['error']['message']}"
  195. except Exception:
  196. error_detail = f"External: {e}"
  197. raise HTTPException(
  198. status_code=r.status_code if r != None else 500,
  199. detail=error_detail,
  200. )
  201. elif app.state.config.TTS_ENGINE == "elevenlabs":
  202. payload = None
  203. try:
  204. payload = json.loads(body.decode("utf-8"))
  205. except Exception as e:
  206. log.exception(e)
  207. raise HTTPException(status_code=400, detail="Invalid JSON payload")
  208. voice_id = payload.get("voice", "")
  209. if voice_id not in get_available_voices():
  210. raise HTTPException(
  211. status_code=400,
  212. detail="Invalid voice id",
  213. )
  214. url = f"https://api.elevenlabs.io/v1/text-to-speech/{voice_id}"
  215. headers = {
  216. "Accept": "audio/mpeg",
  217. "Content-Type": "application/json",
  218. "xi-api-key": app.state.config.TTS_API_KEY,
  219. }
  220. data = {
  221. "text": payload["input"],
  222. "model_id": app.state.config.TTS_MODEL,
  223. "voice_settings": {"stability": 0.5, "similarity_boost": 0.5},
  224. }
  225. try:
  226. r = requests.post(url, json=data, headers=headers)
  227. r.raise_for_status()
  228. # Save the streaming content to a file
  229. with open(file_path, "wb") as f:
  230. for chunk in r.iter_content(chunk_size=8192):
  231. f.write(chunk)
  232. with open(file_body_path, "w") as f:
  233. json.dump(json.loads(body.decode("utf-8")), f)
  234. # Return the saved file
  235. return FileResponse(file_path)
  236. except Exception as e:
  237. log.exception(e)
  238. error_detail = "Open WebUI: Server Connection Error"
  239. if r is not None:
  240. try:
  241. res = r.json()
  242. if "error" in res:
  243. error_detail = f"External: {res['error']['message']}"
  244. except Exception:
  245. error_detail = f"External: {e}"
  246. raise HTTPException(
  247. status_code=r.status_code if r != None else 500,
  248. detail=error_detail,
  249. )
  250. @app.post("/transcriptions")
  251. def transcribe(
  252. file: UploadFile = File(...),
  253. user=Depends(get_current_user),
  254. ):
  255. log.info(f"file.content_type: {file.content_type}")
  256. if file.content_type not in ["audio/mpeg", "audio/wav"]:
  257. raise HTTPException(
  258. status_code=status.HTTP_400_BAD_REQUEST,
  259. detail=ERROR_MESSAGES.FILE_NOT_SUPPORTED,
  260. )
  261. try:
  262. ext = file.filename.split(".")[-1]
  263. id = uuid.uuid4()
  264. filename = f"{id}.{ext}"
  265. file_dir = f"{CACHE_DIR}/audio/transcriptions"
  266. os.makedirs(file_dir, exist_ok=True)
  267. file_path = f"{file_dir}/{filename}"
  268. print(filename)
  269. contents = file.file.read()
  270. with open(file_path, "wb") as f:
  271. f.write(contents)
  272. f.close()
  273. if app.state.config.STT_ENGINE == "":
  274. from faster_whisper import WhisperModel
  275. whisper_kwargs = {
  276. "model_size_or_path": WHISPER_MODEL,
  277. "device": whisper_device_type,
  278. "compute_type": "int8",
  279. "download_root": WHISPER_MODEL_DIR,
  280. "local_files_only": not WHISPER_MODEL_AUTO_UPDATE,
  281. }
  282. log.debug(f"whisper_kwargs: {whisper_kwargs}")
  283. try:
  284. model = WhisperModel(**whisper_kwargs)
  285. except Exception:
  286. log.warning(
  287. "WhisperModel initialization failed, attempting download with local_files_only=False"
  288. )
  289. whisper_kwargs["local_files_only"] = False
  290. model = WhisperModel(**whisper_kwargs)
  291. segments, info = model.transcribe(file_path, beam_size=5)
  292. log.info(
  293. "Detected language '%s' with probability %f"
  294. % (info.language, info.language_probability)
  295. )
  296. transcript = "".join([segment.text for segment in list(segments)])
  297. data = {"text": transcript.strip()}
  298. # save the transcript to a json file
  299. transcript_file = f"{file_dir}/{id}.json"
  300. with open(transcript_file, "w") as f:
  301. json.dump(data, f)
  302. print(data)
  303. return data
  304. elif app.state.config.STT_ENGINE == "openai":
  305. if is_mp4_audio(file_path):
  306. print("is_mp4_audio")
  307. os.rename(file_path, file_path.replace(".wav", ".mp4"))
  308. # Convert MP4 audio file to WAV format
  309. convert_mp4_to_wav(file_path.replace(".wav", ".mp4"), file_path)
  310. headers = {"Authorization": f"Bearer {app.state.config.STT_OPENAI_API_KEY}"}
  311. files = {"file": (filename, open(file_path, "rb"))}
  312. data = {"model": app.state.config.STT_MODEL}
  313. print(files, data)
  314. r = None
  315. try:
  316. r = requests.post(
  317. url=f"{app.state.config.STT_OPENAI_API_BASE_URL}/audio/transcriptions",
  318. headers=headers,
  319. files=files,
  320. data=data,
  321. )
  322. r.raise_for_status()
  323. data = r.json()
  324. # save the transcript to a json file
  325. transcript_file = f"{file_dir}/{id}.json"
  326. with open(transcript_file, "w") as f:
  327. json.dump(data, f)
  328. print(data)
  329. return data
  330. except Exception as e:
  331. log.exception(e)
  332. error_detail = "Open WebUI: Server Connection Error"
  333. if r is not None:
  334. try:
  335. res = r.json()
  336. if "error" in res:
  337. error_detail = f"External: {res['error']['message']}"
  338. except Exception:
  339. error_detail = f"External: {e}"
  340. raise HTTPException(
  341. status_code=r.status_code if r != None else 500,
  342. detail=error_detail,
  343. )
  344. except Exception as e:
  345. log.exception(e)
  346. raise HTTPException(
  347. status_code=status.HTTP_400_BAD_REQUEST,
  348. detail=ERROR_MESSAGES.DEFAULT(e),
  349. )
  350. def get_available_models() -> list[dict]:
  351. if app.state.config.TTS_ENGINE == "openai":
  352. return [{"id": "tts-1"}, {"id": "tts-1-hd"}]
  353. elif app.state.config.TTS_ENGINE == "elevenlabs":
  354. headers = {
  355. "xi-api-key": app.state.config.TTS_API_KEY,
  356. "Content-Type": "application/json",
  357. }
  358. try:
  359. response = requests.get(
  360. "https://api.elevenlabs.io/v1/models", headers=headers
  361. )
  362. response.raise_for_status()
  363. models = response.json()
  364. return [
  365. {"name": model["name"], "id": model["model_id"]} for model in models
  366. ]
  367. except requests.RequestException as e:
  368. log.error(f"Error fetching voices: {str(e)}")
  369. return []
  370. @app.get("/models")
  371. async def get_models(user=Depends(get_verified_user)):
  372. return {"models": get_available_models()}
  373. def get_available_voices() -> dict:
  374. """Returns {voice_id: voice_name} dict"""
  375. ret = {}
  376. if app.state.config.TTS_ENGINE == "openai":
  377. ret = {
  378. "alloy": "alloy",
  379. "echo": "echo",
  380. "fable": "fable",
  381. "onyx": "onyx",
  382. "nova": "nova",
  383. "shimmer": "shimmer",
  384. }
  385. elif app.state.config.TTS_ENGINE == "elevenlabs":
  386. try:
  387. ret = get_elevenlabs_voices()
  388. except Exception:
  389. # Avoided @lru_cache with exception
  390. pass
  391. return ret
  392. @lru_cache
  393. def get_elevenlabs_voices() -> dict:
  394. """
  395. Note, set the following in your .env file to use Elevenlabs:
  396. AUDIO_TTS_ENGINE=elevenlabs
  397. AUDIO_TTS_API_KEY=sk_... # Your Elevenlabs API key
  398. AUDIO_TTS_VOICE=EXAVITQu4vr4xnSDxMaL # From https://api.elevenlabs.io/v1/voices
  399. AUDIO_TTS_MODEL=eleven_multilingual_v2
  400. """
  401. headers = {
  402. "xi-api-key": app.state.config.TTS_API_KEY,
  403. "Content-Type": "application/json",
  404. }
  405. try:
  406. # TODO: Add retries
  407. response = requests.get("https://api.elevenlabs.io/v1/voices", headers=headers)
  408. response.raise_for_status()
  409. voices_data = response.json()
  410. voices = {}
  411. for voice in voices_data.get("voices", []):
  412. voices[voice["voice_id"]] = voice["name"]
  413. except requests.RequestException as e:
  414. # Avoid @lru_cache with exception
  415. log.error(f"Error fetching voices: {str(e)}")
  416. raise RuntimeError(f"Error fetching voices: {str(e)}")
  417. return voices
  418. @app.get("/voices")
  419. async def get_voices(user=Depends(get_verified_user)):
  420. return {"voices": [{"id": k, "name": v} for k, v in get_available_voices().items()]}