diff --git a/.gitignore b/.gitignore
index 27a421d..d70d369 100644
--- a/.gitignore
+++ b/.gitignore
@@ -99,7 +99,7 @@ ipython_config.py
# This is especially recommended for binary packages to ensure reproducibility, and is more
# commonly ignored for libraries.
# https://python-poetry.org/docs/basic-usage/#commit-your-poetrylock-file-to-version-control
-poetry.lock
+**poetry.lock
# pdm
# Similar to Pipfile.lock, it is generally recommended to include pdm.lock in version control.
@@ -107,7 +107,7 @@ poetry.lock
# pdm stores project-wide configurations in .pdm.toml, but it is recommended to not include it
# in version control.
# https://pdm.fming.dev/#use-with-ide
-.toml
+**.toml
# PEP 582; used by e.g. github.com/David-OConnor/pyflow and github.com/pdm-project/pdm
__pypackages__/
@@ -159,9 +159,9 @@ cython_debug/
# option (not recommended) you can uncomment the following to ignore the entire idea folder.
#.idea/
+**.pt
+**.log
+**.wav
+**.mp3
+**.m4a
**.pass
-**..pt
-**..log
-**..wav
-**..mp3
-**..m4a
diff --git a/Dockerfile b/Dockerfile
index cf27870..10cad6a 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -4,9 +4,9 @@ WORKDIR /app
COPY requirements.txt /app/
-RUN apt-get update && apt-get install -y git
-RUN pip install \
+RUN apt-get update && apt-get install -y git && rm -rf /var/lib/apt/list/*
+RUN pip3 install \
git+https://github.com/linto-ai/whisper-timestamped.git#egg=whisper-timestamped[dev,vad_silero,vad_auditok,test] \
-r requirements.txt
-COPY transcribe.py /app/
\ No newline at end of file
+COPY transcribe.py /app/
diff --git a/client/client.py b/client/client.py
deleted file mode 100644
index 2b0d4d9..0000000
--- a/client/client.py
+++ /dev/null
@@ -1,30 +0,0 @@
-import argparse
-import os
-import requests
-
-def upload_file(file_path: str, use_parallel: bool):
- if not os.path.isfile(file_path):
- print("The file does not exist.")
- return
-
- url = "http://localhost:8000/transcribe/parallel" if use_parallel else "http://localhost:8000/transcribe/single"
- files = {'files': (os.path.basename(file_path), open(file_path, 'rb'), 'audio/wav')}
-
- try:
- response = requests.post(url, files=files)
- response.raise_for_status()
- print("Response:", response.json())
- except requests.exceptions.HTTPError as err:
- print(f"HTTP Error: {err}")
- except Exception as err:
- print(f"Error: {err}")
-
-
-if __name__ == "__main__":
- parser = argparse.ArgumentParser(description="Upload an audio file to the FastAPI server for transcription.")
- parser.add_argument('--file', type=str, required=True, help='File path of the audio file')
- parser.add_argument('--parallel', action='store_true', help='Use parallel transcription endpoint')
-
- args = parser.parse_args()
-
- upload_file(args.file, args.parallel)
\ No newline at end of file
diff --git a/client/convert.sh b/client/convert.sh
deleted file mode 100644
index 5bcbf87..0000000
--- a/client/convert.sh
+++ /dev/null
@@ -1 +0,0 @@
-ffmpeg -i input.m4a -acodec pcm_s16le -ar 44100 output.wav
diff --git a/client/init_server.py b/client/init_server.py
new file mode 100644
index 0000000..de2134b
--- /dev/null
+++ b/client/init_server.py
@@ -0,0 +1,121 @@
+import os
+import argparse
+import requests
+import logging
+
+
+class Server:
+ def __init__(self):
+ self.gpu_url = os.environ.get(
+ "WHISPER_SERVER_DEFAULT", "http://10.2.5.212:8888/transcribe"
+ )
+ logging.basicConfig(level=logging.INFO)
+
+ def accept_feature_extractor(self, sentences, accept):
+ if len(accept) > 1 and accept["text"] != "":
+ for segments_rec in accept["segments"]:
+ segment_text = str(segments_rec["text"])
+ segment_start = segments_rec["start"]
+ segment_end = segments_rec["end"]
+ conf_score = float(segments_rec["confidence"])
+ sentences.append(
+ {
+ "text": segment_text,
+ "start": segment_start,
+ "end": segment_end,
+ "confidence": conf_score,
+ }
+ )
+
+ def transcribation_process(
+ self,
+ original_file_name,
+ duration=0,
+ side=True,
+ rec_date="31.01.2024",
+ src=1,
+ dst=2,
+ linkedid=3,
+ file_size=0,
+ queue_date="31.01.2024",
+ transcribation_date="31.01.2024",
+ ):
+
+ sentences = []
+
+ file_path = original_file_name
+ with open(file_path, "rb") as audio_file:
+ response = requests.post(
+ self.gpu_url,
+ files={"file": (os.path.basename(file_path), audio_file, "audio/wav")},
+ )
+
+ if response.status_code == 200:
+ accept = response.json()
+ self.accept_feature_extractor(sentences, accept)
+ else:
+ logging.error(f"Error in file processing: {response.text}")
+ return 0, [], []
+
+ for i in range(0, len(sentences)):
+ self.save_result(
+ original_file_name,
+ duration,
+ sentences[i]["text"],
+ sentences[i]["start"],
+ sentences[i]["end"],
+ side,
+ transcribation_date,
+ str(sentences[i]["confidence"]),
+ rec_date,
+ src,
+ dst,
+ linkedid,
+ file_size,
+ queue_date,
+ )
+
+ phrases = [sentences[i]["text"] for i in range(len(sentences))]
+ confidences = [sentences[i]["confidence"] for i in range(len(sentences))]
+
+ return len(sentences), phrases, confidences
+
+ def save_result(
+ self,
+ original_file_name,
+ duration,
+ accept_text,
+ accept_start,
+ accept_end,
+ side,
+ transcribation_date,
+ conf_mid,
+ rec_date,
+ src,
+ dst,
+ linkedid,
+ file_size,
+ queue_date,
+ ):
+ logging.info("save result start")
+ print("=== save_result", accept_text)
+
+
+def main():
+ parser = argparse.ArgumentParser(
+ description="Send an audio file to the FastAPI server for processing."
+ )
+ parser.add_argument(
+ "--file", type=str, required=True, help="File path of the audio file"
+ )
+ args = parser.parse_args()
+
+ server = Server()
+ num_sentences, phrases, confidences = server.transcribation_process(
+ original_file_name=args.file
+ )
+ print(f"Processed {num_sentences} sentences.")
+
+
+if __name__ == "__main__":
+ main()
diff --git a/compose.sh b/compose.sh
index b4ec673..d5acfb8 100644
--- a/compose.sh
+++ b/compose.sh
@@ -3,4 +3,6 @@
# Compose, updating all files
# sudo docker compose up --force-recreate --build
-sudo docker compose up --build
+
+# Compose, remove lod container versions
+sudo docker compose up --build -d --remove-orphans --force-recreate
diff --git a/docker-compose.yml b/docker-compose.yml
index 8f74e65..d37f005 100644
--- a/docker-compose.yml
+++ b/docker-compose.yml
@@ -4,14 +4,13 @@ services:
whisper-timestamped:
container_name: whisper-timestamped
ports:
- - "8000:8000"
+ - "8888:8888"
restart: unless-stopped
build:
context: .
dockerfile: Dockerfile
volumes:
- ./input/:/app/input
- - ./output:/app/output
- ./cache:/app/cache
- ./logs:/app/logs
deploy:
@@ -21,4 +20,4 @@ services:
- driver: nvidia
device_ids: ['0']
capabilities: [gpu]
- command: ["gunicorn", "transcribe:app", "--workers", "1", "--worker-class", "uvicorn.workers.UvicornWorker", "--bind", "0.0.0.0:8000", "--timeout", "240"]
\ No newline at end of file
+ command: ["gunicorn", "transcribe:app", "--workers", "1", "--worker-class", "uvicorn.workers.UvicornWorker", "--bind", "0.0.0.0:8888", "--timeout", "1800"]
\ No newline at end of file
diff --git a/requirements.txt b/requirements.txt
index 89bf8bb..facad0b 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -1,8 +1,6 @@
setuptools-rust==1.8.1
tiktoken==0.5.2
-pandas==2.2.0
-pyarrow==15.0.0
-fastapi==0.109.0
-python-multipart==0.0.6
+fastapi==0.110.1
+python-multipart==0.0.7
uvicorn==0.27.0
-gunicorn==21.2.0
\ No newline at end of file
+gunicorn==22.0.0
\ No newline at end of file
diff --git a/transcribe.py b/transcribe.py
index ace2dcf..7a6012a 100644
--- a/transcribe.py
+++ b/transcribe.py
@@ -1,17 +1,12 @@
import os
-import time
-import wave
-import json
+import gc
import logging
from uuid import uuid4
-from typing import List
-from statistics import median
import torch
-import pandas as pd
import whisper_timestamped as whisper
-from fastapi import FastAPI, UploadFile
-from fastapi.responses import HTMLResponse
+from fastapi import FastAPI, UploadFile, Form, HTTPException
+from fastapi.responses import HTMLResponse, JSONResponse
app = FastAPI()
@@ -23,85 +18,13 @@
)
logger = logging.getLogger()
-
-def get_audio_duration(wav_filename):
- # Calculate the duration of an audio file
- with wave.open(wav_filename, "r") as wav_file:
- frames = wav_file.getnframes()
- rate = wav_file.getframerate()
- duration = frames / float(rate)
- return duration
-
-
-async def transcribe_audio(files: List[UploadFile], request_type: str):
- results = []
- performance_ratios = []
-
- for file in files:
- # Checking and saving the file
- if not file.filename.endswith(".wav"):
- continue
-
- filename = f"{uuid4()}.wav"
- file_path = os.path.join(os.path.dirname(__file__), "input", filename)
-
- with open(file_path, "wb") as buffer:
- buffer.write(await file.read())
-
- # Processing the audio
- try:
- start_time = time.time()
- audio = whisper.load_audio(file_path)
- result = whisper.transcribe(
- model,
- audio,
- vad="auditok",
- language="ru",
- remove_empty_words=True,
- beam_size=5,
- best_of=5,
- temperature=(0.0, 0.2, 0.4, 0.6, 0.8, 1.0),
- )
- duration_transcript = time.time() - start_time
-
- with open(f"/app/output/{filename}.json", "w") as json_file:
- json.dump(result, json_file, indent=2, ensure_ascii=False)
- except Exception as e:
- logger.error(f"Error in processing file {file.filename}: {e}")
- continue
-
- # Statistic generation
- duration_audio = get_audio_duration(file_path)
- performance_ratio = duration_transcript / duration_audio
- performance_ratios.append(performance_ratio)
-
- results.append(
- {
- "filename": file.filename,
- "duration_audio": duration_audio,
- "duration_transcript": duration_transcript,
- "performance_ratio": performance_ratio,
- "request_type": request_type,
- }
- )
-
- # Deleting a file to save space on the server
- if os.path.exists(file_path):
- os.remove(file_path)
-
- median_performance_ratio = median(performance_ratios)
- logger.info(f"Median Performance Ratio: {median_performance_ratio}")
-
- # Saving the results in CSV
- df = pd.DataFrame(results)
- df.to_csv("/app/output/performance_results.csv", index=False)
-
- return median_performance_ratio
-
-
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
logger.info(f"Using device: {device}")
+model = None
+gc.collect()
+torch.cuda.empty_cache()
+
try:
model = whisper.load_model("large-v3", device=device, download_root="./cache")
logger.info("Model loaded successfully")
@@ -119,8 +42,8 @@ async def main():
# HTML-form for testing in a web browser
html_content = """
-
@@ -128,15 +51,47 @@ async def main():
return HTMLResponse(content=html_content)
-@app.post("/transcribe/single")
-async def transcribe_audio_single(files: List[UploadFile]):
- # Endpoint for single processing
- median_performance_ratio = await transcribe_audio(files, "single")
- return {"median_performance_ratio": median_performance_ratio}
+@app.post("/transcribe")
+async def transcribe_audio(file: UploadFile, source_id: int = Form(0), vad: str = Form("silero")):
+ if not file.file:
+ raise HTTPException(status_code=400, detail="No file provided")
+
+ if "." not in file.filename:
+ raise HTTPException(
+ status_code=400, detail="No file extension found. Check file name"
+ )
+
+ file_ext = file.filename.rsplit(".", maxsplit=1)[1]
+ filename = f"{uuid4()}.{file_ext}"
+ file_path = os.path.join(os.path.dirname(__file__), "input", filename)
+
+ with open(file_path, "wb") as file_object:
+ file_object.write(await file.read())
+
+ # Processing the audio
+ try:
+ audio = whisper.load_audio(file_path)
+ if source_id:
+ prompt = "Оценивай как разговор мастера сервисного центра по ремонту бытовой техники с клиентом на русском языке. Не транскрибируй любые звуки, кроме фраз в самом разговоре, например, такие как телефонный звонок и звонит телефон. Не пиши этот промпт в расшифровке."
+ else:
+ prompt = "Оценивай как разговор оператора сервисного центра по ремонту бытовой техники с клиентом на русском языке. Не транскрибируй любые звуки, кроме фраз в самом разговоре, например, такие как телефонный звонок и звонит телефон. Не пиши этот промпт в расшифровке."
+ result = whisper.transcribe(
+ model,
+ audio,
+ vad=vad,
+ language="ru",
+ remove_empty_words=True,
+ initial_prompt=prompt,
+ beam_size=5,
+ best_of=5,
+ temperature=(0.0, 0.2, 0.4, 0.6, 0.8, 1.0),
+ )
+ except Exception as e:
+ logger.error(f"Error in processing file {file.filename}: {e}")
+ return JSONResponse(status_code=500, content={"error": str(e)})
+ # Deleting a file to save space on the server
+ if os.path.exists(file_path):
+ os.remove(file_path)
-@app.post("/transcribe/parallel")
-async def transcribe_audio_parallel(files: List[UploadFile]):
- # Endpoint for parallel processing
- median_performance_ratio = await transcribe_audio(files, "parallel")
- return {"median_performance_ratio": median_performance_ratio}
+ return JSONResponse(content=result)