EchoSophia

From speech to wisdom

---

Rationale

Echo (Ἠχώ) was a mythological Greek Oread (Ὀρειάς), a mountain nymph and companion of Artemis, the goddess of hunting, wildnerness and wild animals. In his Metamorphoses, Ovid talks about the Jealousy of Hera over her husband Zeus's many different affairs. Whenever she was about to catch him, Echo would distract her with long conversations. When she learned about this, she cursed Echo, the talkative nymph, that she would only be able to repeat the most recently spoken words of another person.

Sophia (Σοφία), is the Greek personification of wisdom. A practical wisdom of sound judgement and intelligence. Distinct from the understanding of Logos (lɒɡɒs), excellence of character and good judgement of Phronesis (φρόνησις) and the spritual knowledge and insight of Gnosis (γνῶσις). Within the Church, the feminine personification of divine wisdom, Ἁγία Σοφία (Hagía Sophía), referred to Christ, the Word of God or to the Holy Spirit, like the dedication of the Eastern Roman empire's Church, in Constantinople.

The culmination of these two concepts, the loquaciousness of Echo with the practical wisdom of Sophia, is the fundamental rationale behind building this tool. Large Language Models have been trained on a vast subset of human records, a monument to our species' efforts of understanding (Logos), however it remains a facsimile of practical wisdom (Sophia). In Plato's Protagoras, the wisdom and knowledge was stolen from Olympus by Prometheus, to help humanity. Similarly, this tool is being built, to aid the user to use AI to extract knowledge from online sources. However, a word of warning must be heeded, just like Echo (due to the curse by Hera) AI/LLMs, are only capable of repeating and regurgitating human speech, writing and knowledge.

The hero image generated for this project, an owl shield with a pen/sword on a field of mountains, is a tribute to Echo, the pursuit of Sophia and the watchful gaze of wisdon; Athena (Minerva).

Installation

uv venv venv
source venv/bin/activate
uv pip install -e .

Usage

• Download audio from YouTube URL

echosophia download <"YOUTUBE_URL">

• convert downloaded *.webm file to a *.wav file

echosophia waver <"AUDIO_FILE">

• split large (>1 hour) audio files into smaller parts

echosophia chunker <"AUDIO_FILE">

• Transcribe audio *.wav file

echosophia transcribe <"AUDIO_FILE"> # --output transcribed.json
echosophia transcribe <"chunks/"> --chunk # transcribe split chunks of a larger audio file

• extract "wisdom" summary from transcripts

echosophia extract <"TRANSCRIPT.json"> <"OUTPUT.md">

Convert to WAV

ffmpeg -i "<FILENAME>.webm" -acodec pcm_s16le -ar 44100 -ac 1 "<filename>.wav"

Split to chunks

ffmpeg -i "<FILENAME>.wav" \
       -f segment \
       -segment_time 00:30:00 \
       -reset_timestamps 1 \
       "<FILENAME_PART>_chunk_%03d.wav"

Chunk validation

for f in chunks/<DATE>/<ROOM>/europython_2026_<ROOM>_<DATE>_chunk_*.wav; do echo -n "$f: "; ffprobe -v error -show_entrie
s format=duration -of csv=p=0 "$f"; done

Development tasks

TODO

References

• MistralAI Le Chat was used as an assistant to lookup API and SDK usage in the docs and help guide the programming when encountering an error.

• The hero image/logo was generated using MistralAI's LeChat.

• The extract_wisdom prompt was modified from the original in danielmiessler's Fabric, an open-source framework for augmenting humans using AI.