Modelpublic

Hochien/wav2vec2-alignment

sourceHugging Faceapache-2.0updated 7mo agoView on Hugging Face

1likes

Model Card

Wav2Vec2-LV-60-Espeak-CV-FT (ONNX)

This is an ONNX export of the facebook/wav2vec2-lv-60-espeak-cv-ft model.

It is designed for client-side inference in the UltrClick ContentPro application to perform forced alignment of lyrics to audio.

Model Details

—Original Model: facebook/wav2vec2-lv-60-espeak-cv-ft
—Format: ONNX (Open Neural Network Exchange)
—Precision: FP16 (Float16)
—Output: IPA Phoneme logits (392 vocab size)
—Sample Rate: 16kHz

Usage

This model is intended to be used with the ONNX Runtime (e.g., via ort in Rust or onnxruntime in Python).

Input

—Name: audio
—Shape: [batch_size, samples]
—Type: Float32 tensor

Output

—Name: logits
—Shape: [batch_size, frames, 392] (392 is the vocab size)

License

This model is a derivative of the original facebook/wav2vec2-lv-60-espeak-cv-ft model and retains the Apache 2.0 license.