5e2094f2ad
FastAPI service in infra/media-worker implementing every MediaInferenceProvider operation behind HTTP API v1, plus capabilities. Bearer auth, MEDIA_ROOT realpath enforcement, strict request schemas, cancellation on disconnect, GPU/CPU gates with 503 overflow, input size/duration limits, lazy SHA256-pinned model downloads, fail-closed redaction with atomic output, offline pytest suite with fake backends, CPU/CUDA/ROCm Dockerfile and a license review of every model. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_0138YqDj3ttfayRwuMKKrqVj
58 lines
2.1 KiB
Python
58 lines
2.1 KiB
Python
"""transcribe: faster-whisper with word timestamps."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import os
|
|
import threading
|
|
from typing import Any
|
|
|
|
from ..errors import BadRequest, Cancelled
|
|
from ..schemas import TranscribeRequest
|
|
from .common import Ctx, clamp01, open_input
|
|
from .devices import ctranslate2_device
|
|
|
|
|
|
class WhisperTranscriber:
|
|
def __init__(self, model_dir: str, settings: Any) -> None:
|
|
from faster_whisper import WhisperModel
|
|
|
|
device = ctranslate2_device(settings)
|
|
self.model = WhisperModel(
|
|
model_dir,
|
|
device=device,
|
|
compute_type="float16" if device == "cuda" else "int8",
|
|
)
|
|
|
|
def transcribe(
|
|
self, path: str, language: str | None, cancel: threading.Event
|
|
) -> tuple[str | None, list[dict[str, Any]]]:
|
|
segments, info = self.model.transcribe(path, language=language, word_timestamps=True, vad_filter=True)
|
|
words: list[dict[str, Any]] = []
|
|
for segment in segments:
|
|
if cancel.is_set():
|
|
raise Cancelled()
|
|
for word in segment.words or []:
|
|
start = int(round(word.start * 1000))
|
|
words.append(
|
|
{
|
|
"startMs": start,
|
|
"endMs": max(int(round(word.end * 1000)), start),
|
|
"text": word.word.strip(),
|
|
"confidence": clamp01(word.probability),
|
|
}
|
|
)
|
|
return info.language, words
|
|
|
|
|
|
def load_transcriber(files: dict[str, str], settings: Any) -> WhisperTranscriber:
|
|
return WhisperTranscriber(os.path.dirname(files["model.bin"]), settings)
|
|
|
|
|
|
def run(ctx: Ctx, body: TranscribeRequest, cancel: threading.Event) -> dict[str, Any]:
|
|
real, info = open_input(ctx, body.path, cancel, need_video=False)
|
|
if not info.has_audio:
|
|
raise BadRequest("input has no audio stream")
|
|
transcriber = ctx.registry.runtime("transcribe")
|
|
language, words = transcriber.transcribe(real, body.language, cancel)
|
|
return {"language": language, "words": [w for w in words if w["text"]]}
|