meetwave-stream (Python)

Live text while people speak (English, Spanish, Portuguese, Russian; German, French and Italian as a preview), and the full transcript (with speakers and names on those tiers, in any language) right after you stop.

pip install meetwave-stream

Python 3.10+; the only dependency is websockets.

import asyncio
import os

import meetwave_stream


def show(final):
    print(final["start"], final["text"])


async def main(chunks):
    session = meetwave_stream.connect(api_key=os.environ["MEETWAVE_API_KEY"], tier="text", language="en")
    session.on("final", show)
    for pcm in chunks:  # 16 kHz mono 16-bit little-endian PCM, any chunk size
        session.send(pcm)
    result = await session.stop()
    print(result.transcript["text"])

• Send audio as soon as you have it. The first connection after a quiet period can take up to 15 s; the client keeps every word (up to 60 s) and sends it when the server is ready.
• A dropped connection is resumed on its own (6 attempts, 0.5 s to 8 s apart) from the last audio the server stored. Nothing is billed twice and no final is repeated.
• Events: ready, partial, final, warning (low balance), reconnecting, error, closed.
• A long phrase without a pause is cut at about 20 s, sometimes inside a word. A final (or partial) that continues the previous final's last word carries "joins_previous": True; meetwave_stream.append_text(text, final) joins it with no space.
• Call send, stop and close from the event loop that runs the session.

Tiers and languages

Every tier has the same live text. It depends on language:

The client names the language in the connection address too (.../v1/stream/ru), so the first text comes sooner.

Outside English, live text writes numbers as words; the transcript at stop has digits. Billing is per second of audio at the tier's price.

Docs: https://developers.meetwave.io/docs/streaming-sdk