In this post, we walk through building a scalable, event-driven transcription pipeline that automatically processes audio files uploaded to Amazon Simple Storage Service (Amazon S3), and show you how to use Amazon EC2 Spot Instances and buffered streaming inference to further reduce costs.

Parakeet-TDT-0.6B-v3, released in August 2025, is an open-source multilingual ASR model that delivers high accuracy across 25 European languages with automatic language detection and flexible licensing under CC-BY-4.0. According to NVIDIA’s published metrics, the model maintains a 6.34% word error rate (WER) in clean conditions and 11.66% WER at 0 dB SNR, and supports audio up to three hours using local attention mode.

The 25 supported languages include Bulgarian, Croatian, Czech, Danish, Dutch, English, Estonian, Finnish, French, German, Greek, Hungarian, Italian, Latvian, Lithuanian, Maltese, Polish, Portuguese, Romanian, Slovak, Slovenian, Spanish, Swedish, Russian, and Ukrainian. This can help alleviate the need for separate models or language-specific configuration when serving international European economies. For deployment on AWS, the model requires GPU-enabled instances with a minimum of 4 GB VRAM, though 8 GB provides better performance. G6 instances (NVIDIA L4 GPUs) provide the best cost-to-performance ratio for inference workloads based on our tests. The model also performs well on G5 (A10G), G4dn (T4), and for maximum throughput, P5 (H100) or P4 (A100) instances.

The process begins when you upload an audio file to an S3 bucket. This triggers an Amazon EventBridge rule that submits a job to AWS Batch. AWS Batch provisions GPU-accelerated compute resources, and the provisioned instances pull our container image with a pre-cached model from Amazon Elastic Container Registry (Amazon ECR). The inference script downloads and processes the file, then uploads the timestamped JSON transcript to an output S3 bucket. The architecture scales to zero when idle, so costs are incurred only during active compute.