🗣️ Speech-to-Text API (Whisper)

Speech-to-Text API is a microservice based on FastAPI and Whisper (OpenAI) that transcribes audio and video into text.
The service supports 99+ languages, including English, Russian, Chinese, Spanish, and many more.

🚀 Features

• 🎤 Audio transcription (WAV, MP3, FLAC, AAC, OGG, M4A, etc.)
• 🎥 Extract audio from video (MP4, MKV, AVI)
• 🌍 Supports 99+ languages (see the table below)
• 🔄 Optional translation into English
• 🏗 Asynchronous API for integration
• 🐳 Docker-ready deployment

---

🌍 Supported Languages

#	Language	Code	#	Language	Code
1	English	en	6	Spanish	es
2	Russian	ru	7	Portuguese	pt
3	Chinese	zh	8	Arabic	ar
4	French	fr	9	Japanese	ja
5	German	de	10	Hindi	hi

Whisper supports 99+ languages. The full list is available in the OpenAI repository.

---

🛠️ Installation & Setup

1️⃣ Run Locally (without Docker)

Requirements: Python 3.9+, FFmpeg

git clone https://github.com/AntonSHBK/speech_to_text_service
cd speech-to-text

# Install dependencies
pip install -r requirements.txt

# Start the FastAPI server
uvicorn app.main:app --host 0.0.0.0 --port 8000

Now the API is available at http://127.0.0.1:8000/docs 📜

---

2️⃣ Run with Docker

Requirements: Docker & Docker Compose

# Build the container
docker-compose build

# Start the service
docker-compose up -d

Now the API is running in a container and available at http://localhost:8000/docs

---

🎯 How to Use the API?

📌 1️⃣ Upload an Audio File

📌 POST /transcribe/ (Send an audio file and receive text)

curl -X 'POST' \
  'http://127.0.0.1:8000/transcribe/' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@audio.mp3'

🔹 Server Response:

{
  "filename": "audio.mp3",
  "transcription": "This is an example transcription"
}

✅ Supported formats: wav, mp3, flac, ogg, aac, m4a, wma, aiff

---

📌 2️⃣ Transcription with Translation to English

📌 Use task="translate"

curl -X 'POST' \
  'http://127.0.0.1:8000/transcribe/' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@audio.mp3' \
  -F 'task=translate'

🔹 Server Response:

{
  "filename": "audio.mp3",
  "transcription": "This is an example transcription in English"
}

---

📌 3️⃣ Customize Transcription with Parameters

You can control how the transcription is generated using various parameters.

🔹 Available parameters:

Parameter	Description	Default
language	Language code (ru, en, fr, etc.)	"ru"
task	"transcribe" (default) or "translate"	"transcribe"
temperature	Controls randomness (0.0 = deterministic)	0.1
max_new_tokens	Max length of transcription output	100
repetition_penalty	Penalizes repeated words	1.2
num_beams	Beam search size (higher = better quality)	1
do_sample	Enables sampling (more variability)	false
top_k	Selects next word from k best options	50
top_p	Nucleus sampling threshold	0.95

📌 Example: More accurate transcription (beam search + repetition penalty)

curl -X 'POST' \
  'http://127.0.0.1:8000/transcribe/' \
  -H 'accept: application/json' \
  -H 'Content-Type: multipart/form-data' \
  -F 'file=@audio.mp3' \
  -F 'language=en' \
  -F 'num_beams=5' \
  -F 'repetition_penalty=1.1'

---

🎯 Use Cases

✅ 1. Transcribing interviews & lectures — Automatic speech-to-text for meetings and podcasts
✅ 2. Voice notes & subtitles — Convert speech into subtitles for videos and recordings
✅ 3. Call transcription — Process phone conversations and customer support calls
✅ 4. Voice commands & chatbots — Integrate into voice assistants and AI chatbots

---

🛠 Support & Contributions

If you have ideas for improvement or found a bug, create an Issue or Pull Request in the project repository.

👨‍💻 Author: Anton Pisarenko
🔗 License: Apache License 2.0

🚀 Start using the Speech-to-Text API today! 🎙️