Automatic Speech Recognition
ASR models can transcribe and translate speech into text. They can be used for AI assistant and Conversational AI. We provide the wrapping code around the ASR models to make them run efficiently in streaming. For the i.MX 95 MPU, we support Neutron NPU acceleration.
ソフトウェア詳細
対応製品
- i.MX8MPLUS: i.MX 8M Plus - Arm® Cortex®-A53、機械学習、ビジョン、マルチメディア、インダストリアルIoT
- i.MX93: i.MX 93アプリケーション・プロセッサ・ファミリ – Arm® Cortex®-A55、MLアクセラレーション、電力効率に優れたMPU
- i.MX95: i.MX 95アプリケーション・プロセッサ・ファミリ：eIQ® Neutron NPUを内蔵した高性能のセーフティ対応プラットフォーム
システム要件
Models supported: Moonshine-tiny, Moonshine-base, Whisper-tiny.en, Whisper-base.en, Whisper-small.en, Whisper-medium.en
Footprint and Profiling
| Model | Model size [parameters] | Weights format | WER* [%] | Cores | Time To First Token [seconds] | Full transcription latency following X seconds of speech [seconds] | Library dependency | ||
|---|---|---|---|---|---|---|---|---|---|
| X = 3s | X = 6s | X = 9s | |||||||
| Moonshine-tiny | 27M | Q8 | 5.92 | 6x Cortex-A55 | 0.15 | 0.38 | 0.64 | 0.85 | ONNX |
| Moonshine-base | 61M | Q8 | 4.06 | 6x Cortex-A55 | 0.29 | 0.64 | 0.99 | 1.63 | ONNX |
| Whisper-tiny.en | 39M | Q8 | 7.30 | 6x Cortex-A55 | 0.17 | 0.37 | 0.56 | 0.74 | ONNX |
| Whisper-base.en | 74M | Q8 | 5.11 | 6x Cortex-A55 | 0.28 | 0.59 | 0.97 | 1.23 | ONNX |
| Whisper-small.en | 244M | Q8 | 3.76 | 6x Cortex-A55 | 0.69 | 1.36 | 2.17 | 2.92 | ONNX |
| Whisper-Medium.en | 769M | Q8 | 3.93 | 6x Cortex-A55 | 2.06 | 3.72 | 5.60 | 8.21 | ONNX |
