Upload 7 files (#5)

- Upload 7 files (def3e8836bed3e8a49ecba7e54776af8b48e83fe)


Co-authored-by: Axel Chiu <Axelisme@users.noreply.huggingface.co>

ASR_model/infer.py

@@ -0,0 +1,41 @@
+from typing import List
+import torch
+import argparse
+import shutil
+import tempfile
+from speechbrain.pretrained import EncoderDecoderASR
+
+
+def asr_model_inference(model: EncoderDecoderASR, audios: List[str]) -> List[str]:
+    """
+    convert input audio to words and return the result
+    """
+    tmp_dir = tempfile.mkdtemp()
+    results = [process_audio(model, audio, tmp_dir) for audio in audios]
+    shutil.rmtree(tmp_dir)
+    return results
+
+def process_audio(model: EncoderDecoderASR, audio: str, savedir:str) -> str:
+    """
+    convert input audio to words and return the result
+    """
+    waveform = model.load_audio(audio, savedir=savedir)
+    # Fake a batch:
+    batch = waveform.unsqueeze(0)
+    rel_length = torch.tensor([1.0])
+    predicted_words, predicted_tokens = model.transcribe_batch(
+        batch, rel_length
+    )
+    return predicted_words[0]
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("-I", dest="audio_file", required=True)
+
+    args = parser.parse_args()
+
+    asr_model = EncoderDecoderASR.from_hparams(
+        source="./inference", hparams_file="hyperparams.yaml", savedir="inference", run_opts={"device": "cpu"})
+
+    print(asr_model_inference(asr_model, [args.audio_file]))

ASR_model/inference/asr.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5533d036f8c2922e4e0246d4543b1936f9b1d80df1e09f624e927f5609e8f75f
+size 126714188

ASR_model/inference/hyperparams.yaml

@@ -0,0 +1,139 @@
+tokenizer: !new:sentencepiece.SentencePieceProcessor
+
+pretrainer: !new:speechbrain.utils.parameter_transfer.Pretrainer
+  loadables:
+    lm: !ref <lm_model>
+    tokenizer: !ref <tokenizer>
+    normalizer: !ref <normalizer>
+    asr: !ref <asr_model>
+
+# Feature parameters
+sample_rate: 16000
+n_fft: 400
+n_mels: 80
+hop_length: 20
+
+compute_features: !new:speechbrain.lobes.features.Fbank
+  sample_rate: !ref <sample_rate>
+  n_fft: !ref <n_fft>
+  n_mels: !ref <n_mels>
+  hop_length: !ref <hop_length>
+
+####################### Model parameters ###########################
+# Transformer
+d_model: 256
+nhead: 4
+num_encoder_layers: 12
+num_decoder_layers: 6
+d_ffn: 2048
+transformer_dropout: 0.1
+activation: !name:torch.nn.GELU
+output_neurons: 5000
+vocab_size: 5000
+
+# Outputs
+blank_index: 0
+label_smoothing: 0.1
+pad_index: 0
+bos_index: 1
+eos_index: 2
+unk_index: 0
+
+# Decoding parameters
+min_decode_ratio: 0.0
+max_decode_ratio: 1.0
+valid_search_interval: 10
+valid_beam_size: 10
+test_beam_size: 10
+ctc_weight_decode: 0.3
+lm_weight: 0.2
+
+############################## models ################################
+CNN: !new:speechbrain.lobes.models.convolution.ConvolutionFrontEnd
+  input_shape: !!python/tuple [8, 10, 8]
+  num_blocks: 2
+  num_layers_per_block: 1
+  out_channels: !!python/tuple [256, 256]
+  kernel_sizes: !!python/tuple [3, 3]
+  strides: !!python/tuple [2, 2]
+  residuals: !!python/tuple [False, False]
+
+Transformer:
+  !new:speechbrain.lobes.models.transformer.TransformerASR.TransformerASR # yamllint disable-line rule:line-length
+  input_size: 5120
+  tgt_vocab: !ref <output_neurons>
+  d_model: !ref <d_model>
+  nhead: !ref <nhead>
+  num_encoder_layers: !ref <num_encoder_layers>
+  num_decoder_layers: !ref <num_decoder_layers>
+  d_ffn: !ref <d_ffn>
+  dropout: !ref <transformer_dropout>
+  activation: !ref <activation>
+  normalize_before: True
+
+lm_model:
+  !new:speechbrain.lobes.models.transformer.TransformerLM.TransformerLM # yamllint disable-line rule:line-length
+  vocab: !ref <output_neurons>
+  d_model: 576
+  nhead: 6
+  num_encoder_layers: 6
+  num_decoder_layers: 0
+  d_ffn: 1538
+  dropout: 0.2
+  activation: !name:torch.nn.GELU
+  normalize_before: False
+
+ctc_lin: !new:speechbrain.nnet.linear.Linear
+  input_size: !ref <d_model>
+  n_neurons: !ref <output_neurons>
+
+seq_lin: !new:speechbrain.nnet.linear.Linear
+  input_size: !ref <d_model>
+  n_neurons: !ref <output_neurons>
+
+encoder: !new:speechbrain.nnet.containers.LengthsCapableSequential
+  input_shape: [null, null, !ref <n_mels>]
+  compute_features: !ref <compute_features>
+  normalize: !ref <normalizer>
+  cnn: !ref <CNN>
+  transformer_encoder: !ref <Tencoder>
+
+asr_model: !new:torch.nn.ModuleList
+  - [!ref <CNN>, !ref <Transformer>, !ref <seq_lin>, !ref <ctc_lin>]
+
+decoder: !new:speechbrain.decoders.S2STransformerBeamSearch
+  modules: [!ref <Transformer>, !ref <seq_lin>, !ref <ctc_lin>]
+  bos_index: !ref <bos_index>
+  eos_index: !ref <eos_index>
+  blank_index: !ref <blank_index>
+  min_decode_ratio: !ref <min_decode_ratio>
+  max_decode_ratio: !ref <max_decode_ratio>
+  beam_size: !ref <test_beam_size>
+  ctc_weight: !ref <ctc_weight_decode>
+  lm_weight: !ref <lm_weight>
+  lm_modules: !ref <lm_model>
+  temperature: 1.15
+  temperature_lm: 1.15
+  using_eos_threshold: False
+  length_normalization: True
+
+Tencoder:
+  !new:speechbrain.lobes.models.transformer.TransformerASR.EncoderWrapper
+  transformer: !ref <Transformer>
+
+normalizer: !new:speechbrain.processing.features.InputNormalization
+  norm_type: global
+  update_until_epoch: 4
+
+
+
+modules:
+  normalizer: !ref <normalizer>
+  encoder: !ref <encoder>
+  decoder: !ref <decoder>
+# define two optimizers here for two-stage training
+
+
+
+log_softmax: !new:torch.nn.LogSoftmax
+  dim: -1

ASR_model/inference/lm.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44e9e1dbc2935debdd681293515d27a3bb93578e8f4e3b7e01ee1d87c47bb10c
+size 104725990

ASR_model/inference/normalizer.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99a037e5582c33311d233f5465436f79de0331230e5ec433b9534a928325de69
+size 1783

ASR_model/inference/tokenizer.ckpt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:448a66ac83788337506bff24c5be0f0ee97d59491af4f92e2341eca40a3b832c
+size 288715