fix decoding

deepspeech/__init__.py

@@ -123,11 +123,7 @@ if not hasattr(paddle, 'cat'):
 
 ########### hcak paddle.Tensor #############
 def item(x: paddle.Tensor):
-    if x.dtype == paddle.fluid.core_avx.VarDesc.VarType.FP32:
+    return x.numpy().item()
-        return float(x)
-    else:
-        raise ValueError("not support")
-
 
 if not hasattr(paddle.Tensor, 'item'):
     logger.warn(

deepspeech/exps/u2/model.py

@@ -381,8 +381,8 @@ class U2Tester(U2Trainer):
             decoding_chunk_size=cfg.decoding_chunk_size,
             num_decoding_left_chunks=cfg.num_decoding_left_chunks,
             simulate_streaming=cfg.simulate_streaming)
-        decode_time = time.time()
+        decode_time = time.time() - start_time
-        
+
         for target, result in zip(target_transcripts, result_transcripts):
             errors, len_ref = errors_func(target, result)
             errors_sum += errors
@@ -392,13 +392,13 @@ class U2Tester(U2Trainer):
                 fout.write(result + "\n")
             logger.info("\nTarget Transcription: %s\nOutput Transcription: %s" %
                         (target, result))
-            logger.info("Current error rate [%s] = %f" %
+            logger.info("One example error rate [%s] = %f" %
                         (cfg.error_rate_type, error_rate_func(target, result)))
 
         return dict(
             errors_sum=errors_sum,
             len_refs=len_refs,
-            num_ins=num_ins, # num examples
+            num_ins=num_ins,  # num examples
             error_rate=errors_sum / len_refs,
             error_rate_type=cfg.error_rate_type,
             num_frames=audio_len.sum().numpy().item(),
@@ -411,6 +411,7 @@ class U2Tester(U2Trainer):
         self.model.eval()
         logger.info(f"Test Total Examples: {len(self.test_loader.dataset)}")
 
+        stride_ms = self.test_loader.dataset.stride_ms
         error_rate_type = None
         errors_sum, len_refs, num_ins = 0.0, 0, 0
         num_frames = 0.0
@@ -424,11 +425,12 @@ class U2Tester(U2Trainer):
                 len_refs += metrics['len_refs']
                 num_ins += metrics['num_ins']
                 error_rate_type = metrics['error_rate_type']
-                logger.info("Error rate [%s] (%d/?) = %f" %
+                rtf = num_time / (num_frames * stride_ms)
-                            (error_rate_type, num_ins, errors_sum / len_refs))
+                logger.info(
+                    "RTF: %f, Error rate [%s] (%d/?) = %f" %
+                    (rtf, error_rate_type, num_ins, errors_sum / len_refs))
 
-        rtf = num_time / (num_frames * self.test_loader.dataset.stride_ms / 1000.0)
+        rtf = num_time / (num_frames * stride_ms)
-        # logging
         msg = "Test: "
         msg += "epoch: {}, ".format(self.epoch)
         msg += "step: {}, ".format(self.iteration)

deepspeech/frontend/featurizer/audio_featurizer.py

@@ -108,7 +108,7 @@ class AudioFeaturizer(object):
     @property
     def stride_ms(self):
         return self._stride_ms
-        
+
     @property
     def feature_size(self):
         """audio feature size"""

deepspeech/frontend/featurizer/speech_featurizer.py

@@ -148,7 +148,7 @@ class SpeechFeaturizer(object):
             float: time(ms)/frame
         """
         return self._audio_featurizer.stride_ms
-    
+
     @property
     def text_feature(self):
         """Return the text feature object.

deepspeech/io/dataset.py

@@ -63,7 +63,7 @@ class ManifestDataset(Dataset):
                 specgram_type='linear',  # 'linear', 'mfcc', 'fbank'
                 feat_dim=0,  # 'mfcc', 'fbank'
                 delta_delta=False,  # 'mfcc', 'fbank'
-                dither=1.0, # feature dither
+                dither=1.0,  # feature dither
                 target_sample_rate=16000,  # target sample rate
                 use_dB_normalization=True,
                 target_dB=-20,
@@ -188,8 +188,7 @@ class ManifestDataset(Dataset):
         super().__init__()
         self._stride_ms = stride_ms
         self._target_sample_rate = target_sample_rate
-        
+
-        
         self._normalizer = FeatureNormalizer(
             mean_std_filepath) if mean_std_filepath else None
         self._augmentation_pipeline = AugmentationPipeline(
@@ -251,7 +250,7 @@ class ManifestDataset(Dataset):
     @property
     def feature_size(self):
         return self._speech_featurizer.feature_size
-    
+
     @property
     def stride_ms(self):
         return self._speech_featurizer.stride_ms

deepspeech/models/u2.py

@@ -49,10 +49,10 @@ from deepspeech.utils.tensor_utils import pad_sequence
 from deepspeech.utils.tensor_utils import th_accuracy
 from deepspeech.utils.utility import log_add
 
-logger = Log(__name__).getlog()
-
 __all__ = ["U2Model", "U2InferModel"]
 
+logger = Log(__name__).getlog()
+
 
 class U2BaseModel(nn.Module):
     """CTC-Attention hybrid Encoder-Decoder model"""
@@ -398,14 +398,17 @@ class U2BaseModel(nn.Module):
         assert decoding_chunk_size != 0
         batch_size = speech.shape[0]
         # Let's assume B = batch_size
+        # encoder_out: (B, maxlen, encoder_dim)
+        # encoder_mask: (B, 1, Tmax)
         encoder_out, encoder_mask = self._forward_encoder(
             speech, speech_lengths, decoding_chunk_size,
-            num_decoding_left_chunks,
+            num_decoding_left_chunks, simulate_streaming)
-            simulate_streaming)  # (B, maxlen, encoder_dim)
         maxlen = encoder_out.size(1)
-        encoder_out_lens = encoder_mask.squeeze(1).sum(1)
+        # (TODO Hui Zhang): bool no support reduce_sum
+        # encoder_out_lens = encoder_mask.squeeze(1).sum(1)
+        encoder_out_lens = encoder_mask.squeeze(1).astype(paddle.int).sum(1)
         ctc_probs = self.ctc.log_softmax(encoder_out)  # (B, maxlen, vocab_size)
-        topk_prob, topk_index = ctc_probs.topk(1, dim=2)  # (B, maxlen, 1)
+        topk_prob, topk_index = ctc_probs.topk(1, axis=2)  # (B, maxlen, 1)
         topk_index = topk_index.view(batch_size, maxlen)  # (B, maxlen)
         pad_mask = make_pad_mask(encoder_out_lens)  # (B, maxlen)
         topk_index = topk_index.masked_fill_(pad_mask, self.eos)  # (B, maxlen)
@@ -573,11 +576,11 @@ class U2BaseModel(nn.Module):
         hyps_lens = hyps_lens + 1  # Add <sos> at begining
         encoder_out = encoder_out.repeat(beam_size, 1, 1)
         encoder_mask = paddle.ones(
-            beam_size, 1, encoder_out.size(1), dtype=paddle.bool)
+            (beam_size, 1, encoder_out.size(1)), dtype=paddle.bool)
         decoder_out, _ = self.decoder(
             encoder_out, encoder_mask, hyps_pad,
             hyps_lens)  # (beam_size, max_hyps_len, vocab_size)
-        decoder_out = paddle.nn.functional.log_softmax(decoder_out, dim=-1)
+        decoder_out = paddle.nn.functional.log_softmax(decoder_out, axis=-1)
         decoder_out = decoder_out.numpy()
         # Only use decoder score for rescoring
         best_score = -float('inf')

deepspeech/utils/tensor_utils.py

@@ -66,7 +66,9 @@ def pad_sequence(sequences: List[paddle.Tensor],
     # assuming trailing dimensions and type of all the Tensors
     # in sequences are same and fetching those from sequences[0]
     max_size = sequences[0].size()
-    trailing_dims = max_size[1:]
+    # (TODO Hui Zhang): slice not supprot `end==start`
+    # trailing_dims = max_size[1:]
+    trailing_dims = max_size[1:] if max_size.ndim >= 2 else ()
     max_len = max([s.size(0) for s in sequences])
     if batch_first:
         out_dims = (len(sequences), max_len) + trailing_dims

examples/aishell/s0/local/data.sh

@@ -1,71 +1,83 @@
 #! /usr/bin/env bash
 
+stage=-1
+stop_stage=100
+
+source ${MAIN_ROOT}/utils/parse_options.sh
+
 mkdir -p data
 TARGET_DIR=${MAIN_ROOT}/examples/dataset
 mkdir -p ${TARGET_DIR}
 
-# download data, generate manifests
+if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
-python3 ${TARGET_DIR}/aishell/aishell.py \
+    # download data, generate manifests
---manifest_prefix="data/manifest" \
+    python3 ${TARGET_DIR}/aishell/aishell.py \
---target_dir="${TARGET_DIR}/aishell"
+    --manifest_prefix="data/manifest" \
+    --target_dir="${TARGET_DIR}/aishell"
+    
+    if [ $? -ne 0 ]; then
+        echo "Prepare Aishell failed. Terminated."
+        exit 1
+    fi
 
-if [ $? -ne 0 ]; then
+    for dataset in train dev test; do
-    echo "Prepare Aishell failed. Terminated."
+        mv data/manifest.${dataset} data/manifest.${dataset}.raw
-    exit 1
+    done
 fi
 
-
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
-for dataset in train dev test; do
+    # download data, generate manifests
-    mv data/manifest.${dataset} data/manifest.${dataset}.raw
+    # build vocabulary
-done
+    python3 ${MAIN_ROOT}/utils/build_vocab.py \
-
+    --unit_type="char" \
-
+    --count_threshold=0 \
-# build vocabulary
+    --vocab_path="data/vocab.txt" \
-python3 ${MAIN_ROOT}/utils/build_vocab.py \
+    --manifest_paths "data/manifest.train.raw"
---unit_type="char" \
+    
---count_threshold=0 \
+    if [ $? -ne 0 ]; then
---vocab_path="data/vocab.txt" \
+        echo "Build vocabulary failed. Terminated."
---manifest_paths "data/manifest.train.raw"
+        exit 1
-
+    fi
-if [ $? -ne 0 ]; then
-    echo "Build vocabulary failed. Terminated."
-    exit 1
 fi
 
 
-# compute mean and stddev for normalizer
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
+    # compute mean and stddev for normalizer
---manifest_path="data/manifest.train.raw" \
+    python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
---specgram_type="fbank" \
+    --manifest_path="data/manifest.train.raw" \
---feat_dim=80 \
+    --specgram_type="fbank" \
---delta_delta=false \
+    --feat_dim=80 \
---stride_ms=10.0 \
+    --delta_delta=false \
---window_ms=25.0 \
+    --stride_ms=10.0 \
---sample_rate=16000 \
+    --window_ms=25.0 \
---num_samples=2000 \
+    --sample_rate=16000 \
---num_workers=0 \
+    --num_samples=-1 \
---output_path="data/mean_std.json"
+    --num_workers=16 \
-
+    --output_path="data/mean_std.json"
-if [ $? -ne 0 ]; then
+    
-    echo "Compute mean and stddev failed. Terminated."
+    if [ $? -ne 0 ]; then
-    exit 1
+        echo "Compute mean and stddev failed. Terminated."
+        exit 1
+    fi
 fi
 
 
-# format manifest with tokenids, vocab size
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-for dataset in train dev test; do
+    # format manifest with tokenids, vocab size
-    python3 ${MAIN_ROOT}/utils/format_data.py \
+    for dataset in train dev test; do
-    --feat_type "raw" \
+        python3 ${MAIN_ROOT}/utils/format_data.py \
-    --cmvn_path "data/mean_std.npz" \
+        --feat_type "raw" \
-    --unit_type "char" \
+        --cmvn_path "data/mean_std.json" \
-    --vocab_path="data/vocab.txt" \
+        --unit_type "char" \
-    --manifest_path="data/manifest.${dataset}.raw" \
+        --vocab_path="data/vocab.txt" \
-    --output_path="data/manifest.${dataset}"
+        --manifest_path="data/manifest.${dataset}.raw" \
-done
+        --output_path="data/manifest.${dataset}"
-
+    done
-if [ $? -ne 0 ]; then
+    
-    echo "Formt mnaifest failed. Terminated."
+    if [ $? -ne 0 ]; then
-    exit 1
+        echo "Formt mnaifest failed. Terminated."
+        exit 1
+    fi
 fi
 
 echo "Aishell data preparation done."

examples/aishell/s1/local/train.sh

@@ -1,6 +1,6 @@
 #! /usr/bin/env bash
 
-ngpu=$(echo ${CUDA_VISIBLE_DEVICES} | python -c 'import sys; a = sys.stdin.read(); print(len(a.split(",")));')
+ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
 echo "using $ngpu gpus..."
 
 python3 -u ${BIN_DIR}/train.py \

examples/tiny/s0/local/data.sh

@@ -1,74 +1,89 @@
 #! /usr/bin/env bash
 
-mkdir -p data
+stage=-1
-TARGET_DIR=${MAIN_ROOT}/examples/dataset
+stop_stage=100
-mkdir -p ${TARGET_DIR}
-
-# download data, generate manifests
-python3 ${TARGET_DIR}/librispeech/librispeech.py \
---manifest_prefix="data/manifest" \
---target_dir="${TARGET_DIR}/librispeech" \
---full_download="False"
-
-if [ $? -ne 0 ]; then
-    echo "Prepare LibriSpeech failed. Terminated."
-    exit 1
-fi
-
-head -n 64 data/manifest.dev-clean  > data/manifest.tiny.raw
 
 # bpemode (unigram or bpe)
 nbpe=200
 bpemode=unigram
 bpeprefix="data/bpe_${bpemode}_${nbpe}"
-# build vocabulary
-python3 ${MAIN_ROOT}/utils/build_vocab.py \
---unit_type "spm" \
---spm_vocab_size=${nbpe} \
---spm_mode ${bpemode} \
---spm_model_prefix ${bpeprefix} \
---vocab_path="data/vocab.txt" \
---manifest_paths="data/manifest.tiny.raw"
 
-if [ $? -ne 0 ]; then
+source ${MAIN_ROOT}/utils/parse_options.sh
-    echo "Build vocabulary failed. Terminated."
-    exit 1
-fi
 
 
-# compute mean and stddev for normalizer
+mkdir -p data
-python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
+TARGET_DIR=${MAIN_ROOT}/examples/dataset
---manifest_path="data/manifest.tiny.raw" \
+mkdir -p ${TARGET_DIR}
---num_samples=64 \
---specgram_type="fbank" \
---feat_dim=80 \
---delta_delta=false \
---sample_rate=16000 \
---stride_ms=10.0 \
---window_ms=25.0 \
---num_workers=0 \
---output_path="data/mean_std.json"
 
-if [ $? -ne 0 ]; then
+if [ ${stage} -le -1 ] && [ ${stop_stage} -ge -1 ]; then
-    echo "Compute mean and stddev failed. Terminated."
+    # download data, generate manifests
-    exit 1
+    python3 ${TARGET_DIR}/librispeech/librispeech.py \
+    --manifest_prefix="data/manifest" \
+    --target_dir="${TARGET_DIR}/librispeech" \
+    --full_download="False"
+    
+    if [ $? -ne 0 ]; then
+        echo "Prepare LibriSpeech failed. Terminated."
+        exit 1
+    fi
+    
+    head -n 64 data/manifest.dev-clean  > data/manifest.tiny.raw
+fi
+
+if [ ${stage} -le 0 ] && [ ${stop_stage} -ge 0 ]; then
+    # build vocabulary
+    python3 ${MAIN_ROOT}/utils/build_vocab.py \
+    --unit_type "spm" \
+    --spm_vocab_size=${nbpe} \
+    --spm_mode ${bpemode} \
+    --spm_model_prefix ${bpeprefix} \
+    --vocab_path="data/vocab.txt" \
+    --manifest_paths="data/manifest.tiny.raw"
+    
+    if [ $? -ne 0 ]; then
+        echo "Build vocabulary failed. Terminated."
+        exit 1
+    fi
 fi
 
 
-# format manifest with tokenids, vocab size
+if [ ${stage} -le 1 ] && [ ${stop_stage} -ge 1 ]; then
-python3 ${MAIN_ROOT}/utils/format_data.py \
+    # compute mean and stddev for normalizer
---feat_type "raw" \
+    python3 ${MAIN_ROOT}/utils/compute_mean_std.py \
---cmvn_path "data/mean_std.npz" \
+    --manifest_path="data/manifest.tiny.raw" \
---unit_type "spm" \
+    --num_samples=64 \
---spm_model_prefix ${bpeprefix} \
+    --specgram_type="fbank" \
---vocab_path="data/vocab.txt" \
+    --feat_dim=80 \
---manifest_path="data/manifest.tiny.raw" \
+    --delta_delta=false \
---output_path="data/manifest.tiny"
+    --sample_rate=16000 \
+    --stride_ms=10.0 \
+    --window_ms=25.0 \
+    --num_workers=2 \
+    --output_path="data/mean_std.json"
+    
+    if [ $? -ne 0 ]; then
+        echo "Compute mean and stddev failed. Terminated."
+        exit 1
+    fi
+fi
 
 
-if [ $? -ne 0 ]; then
+if [ ${stage} -le 2 ] && [ ${stop_stage} -ge 2 ]; then
-    echo "Formt mnaifest failed. Terminated."
+    # format manifest with tokenids, vocab size
-    exit 1
+    python3 ${MAIN_ROOT}/utils/format_data.py \
+    --feat_type "raw" \
+    --cmvn_path "data/mean_std.json" \
+    --unit_type "spm" \
+    --spm_model_prefix ${bpeprefix} \
+    --vocab_path="data/vocab.txt" \
+    --manifest_path="data/manifest.tiny.raw" \
+    --output_path="data/manifest.tiny"
+    
+    
+    if [ $? -ne 0 ]; then
+        echo "Formt mnaifest failed. Terminated."
+        exit 1
+    fi
 fi
 
 echo "LibriSpeech Data preparation done."

examples/tiny/s1/local/test.sh

@@ -1,12 +1,11 @@
 #! /usr/bin/env bash
 
 # download language model
-bash local/download_lm_en.sh
+#bash local/download_lm_en.sh
-if [ $? -ne 0 ]; then
+#if [ $? -ne 0 ]; then
-    exit 1
+#    exit 1
-fi
+#fi
 
-CUDA_VISIBLE_DEVICES=0 \
 python3 -u ${BIN_DIR}/test.py \
 --device 'gpu' \
 --nproc 1 \

examples/tiny/s1/local/train.sh

@@ -1,6 +1,6 @@
 #! /usr/bin/env bash
 
-ngpu=$(echo ${CUDA_VISIBLE_DEVICES} | python -c 'import sys; a = sys.stdin.read(); print(len(a.split(",")));')
+ngpu=$(echo $CUDA_VISIBLE_DEVICES | awk -F "," '{print NF}')
 echo "using $ngpu gpus..."
 
 python3 -u ${BIN_DIR}/train.py \