Merge pull request #8 from ARM-software/add_wav2letter

Added Wav2letter INT8

MODEL_LIST.md

@@ -25,3 +25,8 @@
 | [SSD MobileNet v1 FP32](models/object_detection/ssd_mobilenet_v1/tflite_fp32) | FP32 | TensorFlow Lite |
 | [SSD MobileNet v1 UINT8](models/object_detection/ssd_mobilenet_v1/tflite_uint8) | UINT8 | TensorFlow Lite |
 | [YOLO v3 Tiny FP32](models/object_detection/yolo_v3_tiny/tflite_fp32) | FP32 | TensorFlow Lite |
+
+## Speech Recognition
+| Model Name |  Datatype  |  Framework |
+|-----------------|---------|--------------|
+| [Wav2letter INT8](models/speech_recognition/wav2letter/tflite_int8) | INT8 | TensorFlow Lite |

models/speech_recognition/wav2letter/tflite_int8/README.md

@@ -0,0 +1,49 @@
+# Wav2letter INT8
+
+## Description
+Wav2letter is a convolutional speech recognition neural network. This implementation was created by Arm and quantized to the INT8 datatype.
+
+## License
+[Apache-2.0](https://spdx.org/licenses/Apache-2.0.html)
+
+## Network Information
+| Network Information |  Value         |
+|---------------------|------------------|
+|  Framework          | TensorFlow Lite |
+|  SHA-1 Hash         | 481b7621801363b64dca2cc02b661b26866af76c |
+|  Size (Bytes)       | 23815520 |
+|  Provenance         | https://github.com/ARM-software/ML-zoo/tree/master/models/speech_recognition/wav2letter/tflite_int8 |
+|  Paper              | https://arxiv.org/abs/1609.03193 |
+
+## Accuracy
+Dataset: Librispeech
+
+| Metric | Value |
+|--------|-------|
+| Ler | 0.08771 |
+
+## Performance
+| Platform | Optimized |
+| -------- | ---------- |
+|   CPU    |      :heavy_check_mark:      |
+|   GPU    |      :heavy_check_mark:      |
+
+### Key
+ - :heavy_check_mark: - Optimized for the platform.
+ - :heavy_minus_sign: - Not optimized, but will run on the platform.
+ - :heavy_multiplication_x: - Not optimized and will not run on the platform.
+
+## Optimizations
+| Optimization |  Value  |
+|-----------------|---------|
+| Quantization | INT8 |
+
+## Network Inputs
+| Input Node Name |  Shape  | Description |
+|-----------------|---------|-------------|
+| input_2_int8 | (1, 296, 39) | Speech converted to MFCCs and quantized to INT8. |
+
+## Network Outputs
+| Output Node Name |  Shape  | Description |
+|------------------|---------|-------------|
+| Identity_int8 | (1, 1, 148, 29) | A tensor of time and class probabilities, that represents the probability of each class at each timestep. Should be passed to a decoder. For example ctc_beam_search_decoder. |

models/speech_recognition/wav2letter/tflite_int8/definition.yaml

@@ -0,0 +1,42 @@
+benchmark:
+  LibriSpeech:
+    LER: 0.08771
+description: Wav2letter is a convolutional speech recognition neural network. This
+  implementation was created by Arm and quantized to the INT8 datatype.
+license:
+- Apache-2.0
+network:
+  file_size_bytes: 23815520
+  filename: wav2letter_int8.tflite
+  framework: TensorFlow Lite
+  hash:
+    algorithm: sha1
+    value: 481b7621801363b64dca2cc02b661b26866af76c
+  provenance: https://github.com/ARM-software/ML-zoo/tree/master/models/speech_recognition/wav2letter/tflite_int8
+network_parameters:
+  input_nodes:
+  - description: Speech converted to MFCCs and quantized to INT8.
+    example_input:
+      path: models/speech_recognition/wav2letter/tflite_int8/testing_input/input_2_int8
+    name: input_2_int8
+    shape:
+    - 1
+    - 296
+    - 39
+  output_nodes:
+  - description: A tensor of time and class probabilities, that represents the probability
+      of each class at each timestep. Should be passed to a decoder. For example ctc_beam_search_decoder.
+    name: Identity_int8
+    shape:
+    - 1
+    - 1
+    - 148
+    - 29
+    test_output_path: models/speech_recognition/wav2letter/tflite_int8/testing_output/Identity_int8
+operators:
+  TensorFlow Lite:
+  - CONV_2D
+  - LEAKY_RELU
+  - RESHAPE
+  - SOFTMAX
+paper: https://arxiv.org/abs/1609.03193