Working with audio

Recorded audio

Perigee-1-transcribe accepts WAV, WAVE, MP3, MPEG, OGG, OGA, OPUS, FLAC, AAC, M4A, WMA, AMR, and 3GP uploads up to 200 MB.

The server normalizes accepted recordings for transcription. You do not need to convert a normal supported file to PCM before uploading it.

transcript = client.transcribe(
    file_path="meeting.m4a",
    language="swahili",
)

Realtime audio

Realtime connections receive raw binary frames with these requirements:

The sample_rate sent in the configuration must match the audio source.

The Python SDK's optional audio tools convert common files:

pip install "orbitalsai[audio]"

from orbitalsai.streaming import AudioConverter
 
audio_bytes, sample_rate = AudioConverter.from_file(
    "meeting.mp3",
    target_sample_rate=16000,
)
chunks = AudioConverter.split_chunks(audio_bytes, chunk_size=8000)

Choose recorded or realtime

Use recorded transcription when you have a complete file, want asynchronous processing, or need SRT output. Use realtime transcription when you need partial results, final segments while speech is happening, or word-level timestamps on realtime finals.

Next steps

• Recorded transcription
• Realtime transcription
• Python realtime SDK