babel-scribe

Audio transcription and text translation CLI. Transcribes audio files and translates text files, automatically selecting the right provider based on the source language.

Installation

uv pip install -e .

Environment Variables

Variable	Required for
SARVAM_API_KEY	Indian language transcription and translation
OPENAI_API_KEY	Non-Indian language transcription and translation

Usage

Basic transcription (Hindi → English)

babel-scribe recording.mp3 --from hi

Transcribe and translate (Spanish → French)

babel-scribe recording.mp3 --from es --to fr

With timestamps

babel-scribe recording.mp3 --from hi --timestamps

Translate a text file (Hindi → English)

babel-scribe essay.txt --from hi --to en

Multiple files

babel-scribe file1.mp3 file2.mp3 --from ta --to en

JSON output

babel-scribe recording.mp3 --from hi --output-format json

For all options and examples, run babel-scribe --help.

Provider Routing

Transcription (audio files)

• Indian languages → Sarvam AI (saaras:v3). When the target is English, Sarvam translates in a single step.
• All other languages → OpenAI Whisper (whisper-1). When the target is English, the Whisper translations endpoint is used.

Translation (text files and post-transcription)

• Indian↔English → Sarvam AI (sarvam-translate:v1)
• Indian↔non-English → chained via English (Sarvam + OpenAI)
• All other pairs → OpenAI (gpt-5-mini)