Telegram Audio-to-Text Bot (Deepgram)

This bot lets users upload audio (voice notes, audio files) and returns a .txt file containing the transcription using Deepgram.

Features

• Accepts Telegram voice, audio, video_note, and audio document uploads
• Uses Deepgram prerecorded transcription API with smart formatting and punctuation
• Replies with a downloadable .txt file

Setup

1. Python 3.10+

2. Install dependencies:

python -m venv .venv
source .venv/bin/activate  # Windows: .venv\\Scripts\\activate
pip install -r requirements.txt

3. Configure tokens (recommended via env vars or .env):

• TELEGRAM_BOT_TOKEN: Your Telegram bot token
• DEEPGRAM_API_KEY: Your Deepgram API key
• DATABASE_URL (optional): Postgres URL to persist per-user settings

Create .env from the example:

cp .env.example .env
# edit .env and fill values

Alternatively, the bot will also attempt to read info.txt if present with lines:

Bot token: <telegram_token>
Deepgram token: <deepgram_api_key>

4. Run the bot:

python bot.py

Optional: Persist user settings with Postgres (Supabase)

If you want each user's language/model and Text Intelligence preferences to persist across restarts, provide a Postgres connection string via DATABASE_URL (the bot auto-creates the user_settings table):

DATABASE_URL=postgresql://<user>:<password>@<host>:5432/<database>

For Supabase:

• In your project, go to Settings → Database → Connection string, pick "URI".
• Use the host db.<project_ref>.supabase.co and the database password you configured.
• Example: postgresql://postgres:<your-db-password>@db.<project_ref>.supabase.co:5432/postgres

Notes:

• The bot uses a small connection pool and will create the user_settings table if missing.
• If DATABASE_URL is not set, settings are stored in-memory and reset on restart.

Usage

• Send the bot a voice note or audio file
• The bot will reply with a .txt document containing the transcription

Commands

• /language <English|Vietnamese|en|vi> — set the bot's interface language
• /speechlang <English|Vietnamese|en|vi|auto> — set speech recognition language
• /status — show current language/model settings
• /lang <code|auto> — set language (e.g., en-US, vi) or enable auto-detect
• /detect <on|off> — toggle language detection explicitly
• /model <name> — set model (e.g., nova-2). Send without a name to reset default

Text Intelligence

• /analyze <text> — Analyze text (summary, topics, intents, sentiment)
• /anstatus — Show Text Intelligence settings
• /summarize <off|v2> — Summarizer
• /topics <on|off> — Topic detection
• /intents <on|off> — Intent detection
• /sentiment <on|off> — Sentiment analysis
• /anlang <code> — Analysis language (e.g., en, vi)
• You can also upload .txt/.md/.srt/.vtt files to analyze their contents

Admin (for user id 1578783338)

• /admin — list admin commands
• /adminstatus — show DB status and user count
• /adminget [chat_id] — show stored settings for a user (defaults to current chat)
• /adminset <chat_id> <stt|ti>.<field> <value> — update a setting
  • STT fields: stt.language, stt.detect_language (on/off), stt.model
  • TI fields: ti.language, ti.summarize, ti.topics (on/off), ti.intents (on/off), ti.sentiment (on/off)

Tip for Vietnamese (vi)

• On Deepgram v2, some language/model combos may 400. If that happens, try /lang auto. For best results, upgrade to Python 3.10+ and use model nova-2.

Enable Text Intelligence

• Create a Python 3.10+ virtualenv and install Deepgram v3:
  • python3.10 -m venv .venv && source .venv/bin/activate
  • pip install -U pip
  • pip install -U deepgram-sdk>=3
• Optionally update requirements.txt to deepgram-sdk>=3.0.0 if you are moving the whole project to Python 3.10+.

Notes

• Deepgram supports many audio formats (ogg/opus, mp3, m4a, wav, etc.). The bot passes the file bytes with the best-known mimetype.
• For best accuracy, ensure the audio is clear and not overly compressed.
• Do not commit your real tokens. Use env vars or a local .env file.

Troubleshooting

• If you see a message about missing configuration, ensure the env vars are set or .env contains both keys.
• Network connectivity is required for Deepgram to transcribe.
• If transcription is empty, the audio may be silent, too noisy, or unsupported.
• If DB persistence doesn't work, verify DATABASE_URL is correct and your IP/network can reach the Supabase Postgres endpoint.