batch

Batch Transcription

Transcribe multiple audio files in a single batched encoder forward pass for better throughput.

Build & Run

make build
./build/examples/example-batch model.safetensors vocab.txt audio1.wav audio2.wav audio3.wav

How It Works

transcribe_batch() pads all inputs to the same length, runs a single batched encoder forward pass, then decodes each output independently. This is more efficient than sequential transcription, especially on GPU.

Available on Transcriber, TDTTranscriber, the C API (parakeet_transcriber_transcribe_batch), and the CLI (multiple positional audio args).