WhisperTRT

This project optimizes OpenAI Whisper with NVIDIA TensorRT.

When executing the base.en model on NVIDIA Jetson Orin Nano, WhisperTRT runs ~3x faster while consuming only ~60% the memory compared with PyTorch.

WhisperTRT roughly mimics the API of the original Whisper model, making it easy to use.

Check out the performance and usage details below!

Performance

All benchmarks are generated by calling profile_backends.py, processing a 20 second audio clip.

Execution Time

Execution time in seconds to transcribe 20 seconds of speech on Jetson Orin Nano. See profile_backends.py for details.

	whisper	faster_whisper	whisper_trt
tiny.en	1.74 sec	0.85 sec	0.64 sec
base.en	2.55 sec	Unavailable	0.86 sec

Memory Consumption

Memory consumption to transcribe 20 seconds of speech on Jetson Orin Nano. See profile_backends.py for details.

	whisper	faster_whisper	whisper_trt
tiny.en	569 MB	404 MB	488 MB
base.en	666 MB	Unavailable	439 MB

Usage

Python

from whisper_trt import load_trt_model

model = load_trt_model("tiny.en")

result = model.transcribe("speech.wav") # or pass numpy array

print(result['text'])

You can download an example speech file from here or wget https://www.voiptroubleshooter.com/open_speech/american/OSR_us_000_0010_8k.wav -O speech.wav.

You may want to save or load the model to a custom path. To do so, simply initialize the model like this

model = load_trt_model("tiny.en", path="./my_folder/tiny_en_trt.pth")

Transcribe

This script simply runs the model once.

Please note: The first time you call load_model, it takes some time to build the TensorRT engine. After the first run, the model will be cached in the directory ~/.cache/whisper_trt/.

python examples/transcribe.py tiny.en assets/speech.wav --backend whisper_trt

Profile Backend

This scripts measures the latency and process memory when transcribing audio. It includes a warmup run for more accurate timing.

python examples/profile_backend.py tiny.en assets/speech.wav --backend whisper_trt

Backend can be one of "whisper_trt", "whisper", or "faster_whisper".

Live Transcription

This script demonstrates live transcription using a microphone and voice activity detection.

python examples/live_transcription.py tiny.en --backend whisper_trt

See also

• torch2trt - Used to convert PyTorch model to TensorRT and perform inference.
• NanoLLM - Large Language Models targeting NVIDIA Jetson. Perfect for combining with ASR!