Parakeet ASR

• Source: https://github.com/NVIDIA/NeMo
• License: Apache 2.0
• Alternatives: Whisper, Speaches, faster-whisper
• Status: TBD
• Model: NVIDIA Parakeet-TDT 0.6B

GPU-accelerated speech-to-text built on NVIDIA's Parakeet-TDT model.

Features

• High-Quality Transcription: NVIDIA Parakeet-TDT 0.6B model
• REST API: Simple HTTP interface for audio transcription
• GPU Acceleration: Optimized for NVIDIA GPUs, falls back to CPU
• S3 Integration: Transcribe audio files directly from S3/MinIO

API

Health Check

GET /

Transcribe Audio

POST /transcribe

curl -X POST -F "file=@audio.wav" http://localhost:8000/transcribe

Transcribe from S3

POST /transcribe-s3

{
  "bucket": "my-audio-bucket",
  "key": "recordings/meeting.mp4",
  "endpoint_url": "http://minio:9000"
}

Configuration

Variable	Description
TRANSCRIBE_DEVICE	cuda or cpu
S3_ENDPOINT	S3/MinIO URL
S3_ACCESS_KEY	S3 access key
S3_SECRET_KEY	S3 secret key