Biblicus

Make your documents usable by your assistant, then decide later how you will search and retrieve them.

Speech to text extraction

Speech to text extractors transcribe audio items into text artifacts. The raw audio bytes remain unchanged in the corpus.

Available providers

Biblicus supports multiple speech to text providers. Each is an optional dependency.

stt-openai

Speech to text using OpenAI Whisper.

To install:

python -m pip install "biblicus[openai]"

To configure, set OPENAI_API_KEY or add to ~/.biblicus/config.yml:

openai:
  api_key: YOUR_KEY_HERE

Configuration options:

Option Default Description
model whisper-1 OpenAI transcription model
response_format json Response format (json, verbose_json, text, srt, vtt)
language None Language hint
prompt None Prompt to guide transcription

Example:

python -m biblicus extract build --corpus corpora/demo \
  --stage 'stt-openai:model=whisper-1,language=en'

stt-deepgram

Speech to text using Deepgram.

To install:

python -m pip install "biblicus[deepgram]"

To configure, set DEEPGRAM_API_KEY or add to ~/.biblicus/config.yml:

deepgram:
  api_key: YOUR_KEY_HERE

Configuration options:

Option Default Description
model nova-3 Deepgram model (nova-3, nova-2, etc.)
language None Language code hint
punctuate true Add punctuation
smart_format true Apply smart formatting
diarize false Enable speaker diarization
filler_words false Include filler words

Example:

python -m biblicus extract build --corpus corpora/demo \
  --stage 'stt-deepgram:model=nova-3,language=en'

stt-aldea

Speech to text using the Aldea API.

To install:

python -m pip install "biblicus[aldea]"

To configure, set ALDEA_API_KEY or add to ~/.biblicus/config.yml:

aldea:
  api_key: org_YOUR_KEY_HERE

Configuration options:

Option Default Description
language None Language code (BCP-47, e.g. en-US, es)
diarization false Enable speaker diarization
timestamps false Include per-word timestamps

Example:

python -m biblicus extract build --corpus corpora/demo \
  --stage 'stt-aldea:language=en-US'

Choosing a provider

All three providers transcribe audio items. Choose based on your needs: