s2t (Speech-to-Text CLI)

A lightweight CLI tool for transcribing audio (.mp3, .opus, .m4a, .wav, .flac, .ogg, etc.) directly to plain text or clean multi-page PDF documents using DraxonLLMs.

Features

• Multi-Format Support: Handles .mp3, .opus, .m4a, .wav, .flac, .ogg, etc. Auto-normalizes inputs to 16kHz mono WAV via ffmpeg if available.
• Multilingual LLM Audio Models: Powered by audio-native models (gemini-flash, gemini-pro, qwen3.8-omni-flash, mimo-v2.5-asr).
• Flexible Outputs:
  • Direct raw text to stdout
  • Output to plain text file (.txt)
  • Standalone multi-page PDF export (.pdf) with automatic page wrapping and zero external dependencies
• Live Streaming: Stream tokens directly to terminal in real-time (--stream).

Installation

Via pip:

pip install s2t-py

From source:

git clone https://github.com/DraxonV1/Speech2Text.git
cd Speech2Text
pip install -e .

Quickstart

1. Set API Key

Set your DraxonLLMs API key via environment variable:

# Linux / macOS
export DRAXON_API_KEY="drx_your_api_key_here"

# Windows PowerShell
$env:DRAXON_API_KEY="drx_your_api_key_here"

# Windows Command Prompt
set DRAXON_API_KEY=drx_your_api_key_here

(You can also pass it explicitly with --api-key="drx_...")

2. Transcribe Audio

Print raw text directly to terminal:

s2t --input=voice.opus

Output to plain text file:

s2t --input=meeting.mp3 --output=transcript.txt

Output to formatted PDF document:

s2t --input=interview.m4a --output=interview.pdf

Real-time streaming transcription:

s2t --input=speech.wav --stream

CLI Options

usage: s2t [-h] -i INPUT [-o OUTPUT] [--api-key API_KEY] [--api-base API_BASE]
           [--model MODEL] [--stream]

options:
  -h, --help            show this help message and exit
  -i INPUT, --input INPUT
                        Input audio file (.mp3, .opus, .m4a, .wav, etc.)
  -o OUTPUT, --output OUTPUT
                        Output file (.txt or .pdf). If omitted, prints raw text to stdout.
  --api-key API_KEY     DraxonLLMs API key (default: env DRAXON_API_KEY)
  --api-base API_BASE   API base URL (default: https://ai.draxon.one/api/v1)
  --model MODEL         Model name (default: gemini-flash)
  --stream              Stream raw text from model in real time

Supported Models

Refer to models.txt for the current audio model roster:

• gemini-flash (default): High accuracy, low latency, multilingual (English, Hindi, Spanish, etc.)
• gemini-pro: Maximum reasoning and contextual transcription
• qwen3.8-omni-flash: Omni-channel audio model
• mimo-v2.5-asr: Specialized Chinese speech recognition

To select a different model:

s2t --input=audio.mp3 --model=gemini-pro

License

MIT