A CLI tool and Python library that transcribes audio recordings and generates structured meeting summaries using Groq API. Powered by Whisper for speech-to-text and Llama for summarization.

• Speech-to-text transcription via Groq's Whisper models (near-instant on Groq hardware)
• AI-powered summarization -- key points, decisions, and action items extracted automatically
• Automatic chunking for large files (>25 MB free tier limit) with overlapping segments so no words are lost
• Multilingual support -- works with any language Whisper supports (Russian, English, Spanish, French, German, and many more)
• Batch processing -- drop multiple audio files into input/ and process them all at once
• Importable library -- use it in your own Python projects with a clean API

• Python 3.10+
• ffmpeg installed and available on your PATH (download)
• A free Groq API key from console.groq.com

pip install meeting-recap

Python 3.13+ also requires the audioop-lts shim (automatically installed as a dependency).

git clone https://github.com/talhatek/meeting-recap.git
cd meeting-recap
python -m venv venv && venv\Scripts\activate  # Windows
pip install -e .

Create a .env file with your API key:

GROQ_API_KEY=your_api_key_here

Place one or more audio files into the input/ folder, then run:

# Transcribe & summarize (default: English)
meeting-recap
# or
python main.py

# Specify a different language (ISO-639-1 code)
meeting-recap --language ru

# Use a different Whisper model
meeting-recap --model whisper-large-v3

# Full example with all options
meeting-recap \
  --language en \
  --model whisper-large-v3-turbo \
  --summary-model llama-3.3-70b-versatile \
  --input-dir input \
  --output-dir output

flac, mp3, mp4, mpeg, mpga, m4a, ogg, wav, webm

For each audio file, two files are generated in the output/ directory:

output/
  meeting_transcription.txt   # Full transcription text
  meeting_summary.txt         # Structured summary with key points

Install and import in any Python project:

pip install meeting-recap

from meeting_recap import process

result = process("meeting.mp3", api_key="gsk_...")
print(result.transcription)
print(result.summary)

from meeting_recap import transcribe

text = transcribe("meeting.mp3", api_key="gsk_...", language="ru")
print(text)

from meeting_recap import summarize

summary = summarize(text, api_key="gsk_...", language="ru")
print(summary)

from meeting_recap import process, transcribe, summarize, RecapResult

# process() -- full pipeline
result: RecapResult = process(
    audio_path="meeting.mp3",
    api_key="gsk_...",
    language="en",                        # ISO-639-1 language code
    model="whisper-large-v3-turbo",       # Whisper model
    summary_model="llama-3.3-70b-versatile",  # LLM model
    verbose=True,                         # print progress
)
result.audio_path       # pathlib.Path to the source file
result.transcription    # full transcription string
result.summary          # structured summary string

# transcribe() -- speech-to-text only
text = transcribe(
    audio_path="meeting.mp3",
    api_key="gsk_...",
    language="en",
    model="whisper-large-v3-turbo",
    verbose=True,
)

# summarize() -- summarize existing text
summary = summarize(
    text="...",
    api_key="gsk_...",
    language="en",
    model="llama-3.3-70b-versatile",
)

input/*.mp3
    |
    v
[File size check]
    |
    +--> <= 25 MB --> Groq Whisper API --> Transcription
    |
    +--> > 25 MB ---> Split into overlapping chunks (pydub + ffmpeg)
                        |
                        +--> Chunk 1 --> Groq Whisper API --+
                        +--> Chunk 2 --> Groq Whisper API --+--> Merge --> Transcription
                        +--> Chunk N --> Groq Whisper API --+
                                                                    |
                                                                    v
                                                            Groq LLM (summarize)
                                                                    |
                                                                    v
                                                            output/*_summary.txt
                                                            output/*_transcription.txt

Chunking details:

• Files over 25 MB are split into segments that fit within Groq's free tier upload limit
• 5-second overlap between chunks prevents words from being cut off at boundaries
• Chunks are exported as 64 kbps mono MP3 at 16 kHz (optimal for speech recognition)
• Temporary chunk files are automatically cleaned up after processing

Any language supported by OpenAI Whisper. Common examples:

Full list: Whisper supported languages

MIT