tflite_pruned_int8

Wav2letter Pruned INT8

Description

Wav2letter is a convolutional speech recognition neural network. This implementation was created by Arm, pruned to 50% sparisty, fine-tuned and quantized using the TensorFlow Model Optimization Toolkit.

License

Apache-2.0

Related Materials

Class Labels

The class labels associated with this model can be downloaded by running the script get_class_labels.sh.

Model Recreation Code

Code to recreate this model can be found here.

Network Information

Network Information	Value
Framework	TensorFlow Lite
SHA-1 Hash	1771d122ba1ed9354188491e6efbcbd31cc8ba69
Size (Bytes)	23766192
Provenance	https://github.com/ARM-software/ML-zoo/tree/master/models/speech_recognition/wav2letter/tflite_pruned_int8
Paper	https://arxiv.org/abs/1609.03193

Performance

Platform	Optimized
Cortex-A	✔️
Cortex-M	✔️
Mali GPU	✔️
Ethos U	✔️

Key

• ✔️ - Will run on this platform.
• ✖️ - Will not run on this platform.

Accuracy

Dataset: LibriSpeech

Metric	Value
LER	0.0783

Optimizations

Optimization	Value
Quantization	INT8
Sparsity	50%

Network Inputs

Input Node Name	Shape	Description
input_4	(1, 296, 39)	Speech converted to MFCCs and quantized to INT8

Network Outputs

Output Node Name	Shape	Description
Identity	(1, 1, 148, 29)	A tensor of (batch, time, class probabilities) that represents the probability of each class at each timestep. Should be passed to a decoder e.g. ctc_beam_search_decoder.