Merge branch 'espnet:master' into master

chintu619 · Mar 24, 2022 · 39700a0 · 39700a0
2 parents aa706c5 + 14c6350
commit 39700a0
Show file tree

Hide file tree

Showing 29 changed files with 772 additions and 0 deletions.
diff --git a/egs2/README.md b/egs2/README.md
@@ -57,6 +57,7 @@ See: https://espnet.github.io/espnet/espnet2_tutorial.html#recipes-using-espnet2
 | mini_an4                | Mini version of CMU AN4 database for the integration test                               | ASR/TTS/SE              | ENG                  | http://www.speech.cs.cmu.edu/databases/an4/                                                                  |              |
 | mini_librispeech        | Mini version of Librispeech corpus                                                      | DIAR                    | ENG                  | https://openslr.org/31/                                                                                      |              |
 | mls                     | MLS (A large multilingual corpus derived from LibriVox audiobooks)                      | ASR                     | 8 languages          | http://www.openslr.org/94/                                                                                   |              |
+| ms_indic_is18           | Microsoft Speech Corpus (Indian languages)                                              | ASR                     | 3 langs: TEL TAM GUJ | https://msropendata.com/datasets/7230b4b1-912d-400e-be58-f84e0512985e                                        |              |
 | nsc                     | National Speech Corpus                                                                  | ASR                     | ENG-SG               | https://www.imda.gov.sg/programme-listing/digital-services-lab/national-speech-corpus                        |              |
 | open_li52               | Corpus combination with 52 languages(Commonvocie + voxforge)                            | Multilingual ASR        | 52 languages         |                                                                                                             |              |
 | polyphone_swiss_french  | Swiss French Polyphone corpus                                                           | ASR                     | FRA                  | http://catalog.elra.info/en-us/repository/browse/ELRA-S0030_02                                               |              |

diff --git a/egs2/TEMPLATE/asr1/db.sh b/egs2/TEMPLATE/asr1/db.sh
@@ -128,6 +128,7 @@ PRIMEWORDS_CHINESE=downloads
 SEAME=
 BENGALI=downloads
 IWSLT14=
+MS_INDIC_IS18=
 
 # For only CMU TIR environment
 if [[ "$(hostname)" == tir* ]]; then
@@ -162,6 +163,7 @@ if [[ "$(hostname)" == tir* ]]; then
     PRIMEWORDS_CHINESE=/projects/tir5/data/speech_corpora/Primewords_Chinese
     FISHER_CALLHOME_SPANISH=/projects/tir5/data/speech_corpora/fisher_callhome_spanish
     DSING=/projects/tir5/data/speech_corpora/sing_300x30x2
+    MS_INDIC_IS18=/projects/tir6/general/cnariset/corpora/microsoft_speech_corpus_indian_languages
 fi
 
 # For only JHU environment

diff --git a/egs2/ms_indic_18/asr1/README.md b/egs2/ms_indic_18/asr1/README.md
@@ -0,0 +1,94 @@
+<!-- Generated by scripts/utils/show_asr_result.sh -->
+
+# RESULTS
+## Environments
+- date: `Tue Mar 22 13:38:24 EDT 2022`
+- python version: `3.9.5 (default, Jun  4 2021, 12:28:51)  [GCC 7.5.0]`
+- espnet version: `espnet 0.10.7a1`
+- pytorch version: `pytorch 1.8.1+cu111`
+- Git hash: `f91410f712d1287cd6809c5bf26b54c5a40fe314`
+  - Commit date: `Mon Mar 14 22:32:17 2022 -0400`
+- Pretrained model: [espnet/chai_microsoft_indian_langs_te](https://huggingface.co/espnet/chai_microsoft_indian_langs_te)
+
+## Self-supervised learning features [wav2vec2_xlsr, Conformer, utt_mvn](conf/tuning/train_asr_xlsr53_conformer.yaml) with [Transformer-LM](conf/tuning/train_lm_transformer.yaml) and [RNN-LM](conf/tuning/train_lm_rnn.yaml). During inference, all below models use the same [decoding parameters](conf/tuning/decode_asr_transformer.yaml).
+
+### WER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|28413|78.0|19.5|2.5|2.4|24.4|80.1|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te|3040|28413|78.0|19.4|2.6|2.4|24.4|79.7|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|28413|78.0|19.5|2.6|2.5|24.5|79.9|
+
+### CER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|229419|95.6|2.2|2.2|1.6|6.1|80.1|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te|3040|229419|95.6|2.2|2.2|1.6|6.0|79.7|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|229419|95.6|2.1|2.2|1.6|6.0|79.9|
+
+### TER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|146657|92.7|4.7|2.6|1.6|8.9|80.1|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te|3040|146657|92.8|4.7|2.6|1.6|8.9|79.7|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|146657|92.8|4.6|2.6|1.6|8.9|79.9|
+
+
+
+## Self-supervised learning features [wav2vec2_large_ll60k, Transformer, utt_mvn](conf/tuning/train_asr_wav2vec2.yaml) with [Transformer-LM](conf/tuning/train_lm_transformer.yaml) and [RNN-LM](conf/tuning/train_lm_rnn.yaml)
+
+### WER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|28413|77.3|20.3|2.4|2.9|25.6|79.5|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te|3040|28413|77.4|20.1|2.5|2.8|25.3|79.3|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|28413|77.5|20.1|2.4|2.8|25.3|79.6|
+
+### CER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|229419|95.0|2.5|2.5|1.8|6.8|79.5|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te|3040|229419|95.1|2.4|2.5|1.8|6.7|79.3|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|229419|95.1|2.4|2.5|1.7|6.6|79.6|
+
+### TER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|146657|91.8|5.2|3.0|1.8|9.9|79.5|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave/test_te|3040|146657|91.9|5.1|2.9|1.8|9.8|79.3|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave/test_te|3040|146657|92.0|5.1|3.0|1.7|9.7|79.6|
+
+
+
+## Standard ASR model based on [Transformer](conf/tuning/train_asr_transformer.yaml) with [Transformer-LM](conf/tuning/train_lm_transformer.yaml) and [RNN-LM](conf/tuning/train_lm_rnn.yaml)
+
+### WER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave_10best/test_te|3040|28413|75.7|22.0|2.4|3.2|27.6|82.7|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave_10best/test_te|3040|28413|75.9|21.8|2.4|3.1|27.3|82.2|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave_10best/test_te|3040|28413|76.1|21.5|2.4|3.1|27.0|82.6|
+
+### CER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave_10best/test_te|3040|229419|94.7|2.7|2.6|2.0|7.3|82.7|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave_10best/test_te|3040|229419|94.8|2.7|2.6|2.0|7.2|82.2|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave_10best/test_te|3040|229419|94.8|2.6|2.6|2.0|7.1|82.6|
+
+### TER
+
+|dataset|Snt|Wrd|Corr|Sub|Del|Ins|Err|S.Err|
+|---|---|---|---|---|---|---|---|---|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave_10best/test_te|3040|146657|91.1|5.8|3.1|2.0|10.9|82.7|
+|decode_transformer_lm_lm_train_lm_rnn_te_bpe150_valid.loss.best_asr_model_valid.acc.ave_10best/test_te|3040|146657|91.3|5.7|3.1|2.0|10.7|82.2|
+|decode_transformer_lm_lm_train_lm_transformer_te_bpe150_valid.loss.ave_asr_model_valid.acc.ave_10best/test_te|3040|146657|91.4|5.5|3.1|1.9|10.6|82.6|
+
diff --git a/egs2/ms_indic_18/asr1/asr.sh b/egs2/ms_indic_18/asr1/asr.sh
@@ -0,0 +1 @@
+../../TEMPLATE/asr1/asr.sh
diff --git a/egs2/ms_indic_18/asr1/cmd.sh b/egs2/ms_indic_18/asr1/cmd.sh
@@ -0,0 +1,110 @@
+# ====== About run.pl, queue.pl, slurm.pl, and ssh.pl ======
+# Usage: <cmd>.pl [options] JOB=1:<nj> <log> <command...>
+# e.g.
+#   run.pl --mem 4G JOB=1:10 echo.JOB.log echo JOB
+#
+# Options:
+#   --time <time>: Limit the maximum time to execute.
+#   --mem <mem>: Limit the maximum memory usage.
+#   -–max-jobs-run <njob>: Limit the number parallel jobs. This is ignored for non-array jobs.
+#   --num-threads <ngpu>: Specify the number of CPU core.
+#   --gpu <ngpu>: Specify the number of GPU devices.
+#   --config: Change the configuration file from default.
+#
+# "JOB=1:10" is used for "array jobs" and it can control the number of parallel jobs.
+# The left string of "=", i.e. "JOB", is replaced by <N>(Nth job) in the command and the log file name,
+# e.g. "echo JOB" is changed to "echo 3" for the 3rd job and "echo 8" for 8th job respectively.
+# Note that the number must start with a positive number, so you can't use "JOB=0:10" for example.
+#
+# run.pl, queue.pl, slurm.pl, and ssh.pl have unified interface, not depending on its backend.
+# These options are mapping to specific options for each backend and
+# it is configured by "conf/queue.conf" and "conf/slurm.conf" by default.
+# If jobs failed, your configuration might be wrong for your environment.
+#
+#
+# The official documentation for run.pl, queue.pl, slurm.pl, and ssh.pl:
+#   "Parallelization in Kaldi": http://kaldi-asr.org/doc/queue.html
+# =========================================================~
+
+
+# Select the backend used by run.sh from "local", "stdout", "sge", "slurm", or "ssh"
+cmd_backend='local'
+
+# Local machine, without any Job scheduling system
+if [ "${cmd_backend}" = local ]; then
+
+    # The other usage
+    export train_cmd="run.pl"
+    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
+    export cuda_cmd="run.pl"
+    # Used for "*_recog.py"
+    export decode_cmd="run.pl"
+
+# Local machine logging to stdout and log file, without any Job scheduling system
+elif [ "${cmd_backend}" = stdout ]; then
+
+    # The other usage
+    export train_cmd="stdout.pl"
+    # Used for "*_train.py": "--gpu" is appended optionally by run.sh
+    export cuda_cmd="stdout.pl"
+    # Used for "*_recog.py"
+    export decode_cmd="stdout.pl"
+
+
+# "qsub" (Sun Grid Engine, or derivation of it)
+elif [ "${cmd_backend}" = sge ]; then
+    # The default setting is written in conf/queue.conf.
+    # You must change "-q g.q" for the "queue" for your environment.
+    # To know the "queue" names, type "qhost -q"
+    # Note that to use "--gpu *", you have to setup "complex_value" for the system scheduler.
+
+    export train_cmd="queue.pl"
+    export cuda_cmd="queue.pl"
+    export decode_cmd="queue.pl"
+
+
+# "qsub" (Torque/PBS.)
+elif [ "${cmd_backend}" = pbs ]; then
+    # The default setting is written in conf/pbs.conf.
+
+    export train_cmd="pbs.pl"
+    export cuda_cmd="pbs.pl"
+    export decode_cmd="pbs.pl"
+
+
+# "sbatch" (Slurm)
+elif [ "${cmd_backend}" = slurm ]; then
+    # The default setting is written in conf/slurm.conf.
+    # You must change "-p cpu" and "-p gpu" for the "partition" for your environment.
+    # To know the "partion" names, type "sinfo".
+    # You can use "--gpu * " by default for slurm and it is interpreted as "--gres gpu:*"
+    # The devices are allocated exclusively using "${CUDA_VISIBLE_DEVICES}".
+
+    export train_cmd="slurm.pl"
+    export cuda_cmd="slurm.pl"
+    export decode_cmd="slurm.pl"
+
+elif [ "${cmd_backend}" = ssh ]; then
+    # You have to create ".queue/machines" to specify the host to execute jobs.
+    # e.g. .queue/machines
+    #   host1
+    #   host2
+    #   host3
+    # Assuming you can login them without any password, i.e. You have to set ssh keys.
+
+    export train_cmd="ssh.pl"
+    export cuda_cmd="ssh.pl"
+    export decode_cmd="ssh.pl"
+
+# This is an example of specifying several unique options in the JHU CLSP cluster setup.
+# Users can modify/add their own command options according to their cluster environments.
+elif [ "${cmd_backend}" = jhu ]; then
+
+    export train_cmd="queue.pl --mem 2G"
+    export cuda_cmd="queue-freegpu.pl --mem 2G --gpu 1 --config conf/queue.conf"
+    export decode_cmd="queue.pl --mem 4G"
+
+else
+    echo "$0: Error: Unknown cmd_backend=${cmd_backend}" 1>&2
+    return 1
+fi
diff --git a/egs2/ms_indic_18/asr1/conf/decode_asr.yaml b/egs2/ms_indic_18/asr1/conf/decode_asr.yaml
@@ -0,0 +1 @@
+conf/tuning/decode_asr_transformer.yaml
diff --git a/egs2/ms_indic_18/asr1/conf/fbank.conf b/egs2/ms_indic_18/asr1/conf/fbank.conf
@@ -0,0 +1,2 @@
+--sample-frequency=16000 
+--num-mel-bins=80
diff --git a/egs2/ms_indic_18/asr1/conf/pbs.conf b/egs2/ms_indic_18/asr1/conf/pbs.conf
@@ -0,0 +1,11 @@
+# Default configuration
+command qsub -V -v PATH -S /bin/bash
+option name=* -N $0
+option mem=* -l mem=$0
+option mem=0          # Do not add anything to qsub_opts
+option num_threads=* -l ncpus=$0
+option num_threads=1  # Do not add anything to qsub_opts
+option num_nodes=* -l nodes=$0:ppn=1
+default gpu=0
+option gpu=0
+option gpu=* -l ngpus=$0
diff --git a/egs2/ms_indic_18/asr1/conf/pitch.conf b/egs2/ms_indic_18/asr1/conf/pitch.conf
@@ -0,0 +1 @@
+--sample-frequency=16000
diff --git a/egs2/ms_indic_18/asr1/conf/queue.conf b/egs2/ms_indic_18/asr1/conf/queue.conf
@@ -0,0 +1,12 @@
+# Default configuration
+command qsub -v PATH -cwd -S /bin/bash -j y -l arch=*64*
+option name=* -N $0
+option mem=* -l mem_free=$0,ram_free=$0
+option mem=0          # Do not add anything to qsub_opts
+option num_threads=* -pe smp $0
+option num_threads=1  # Do not add anything to qsub_opts
+option max_jobs_run=* -tc $0
+option num_nodes=* -pe mpi $0  # You must set this PE as allocation_rule=1
+default gpu=0
+option gpu=0
+option gpu=* -l gpu=$0 -q g.q
diff --git a/egs2/ms_indic_18/asr1/conf/slurm.conf b/egs2/ms_indic_18/asr1/conf/slurm.conf
@@ -0,0 +1,14 @@
+# Default configuration
+command sbatch --export=PATH
+option name=* --job-name $0
+option time=* --time $0
+option mem=* --mem-per-cpu $0
+option mem=0
+option num_threads=* --cpus-per-task $0
+option num_threads=1 --cpus-per-task 1
+option num_nodes=* --nodes $0
+default gpu=0
+option gpu=0 -p cpu
+option gpu=* -p gpu --gres=gpu:$0 -c $0  # Recommend allocating more CPU than, or equal to the number of GPU
+# note: the --max-jobs-run option is supported as a special case
+# by slurm.pl and you don't have to handle it in the config file.
diff --git a/egs2/ms_indic_18/asr1/conf/train_asr.yaml b/egs2/ms_indic_18/asr1/conf/train_asr.yaml
@@ -0,0 +1 @@
+conf/tuning/train_asr_transformer.yaml
diff --git a/egs2/ms_indic_18/asr1/conf/train_lm.yaml b/egs2/ms_indic_18/asr1/conf/train_lm.yaml
@@ -0,0 +1 @@
+conf/tuning/train_lm_rnn.yaml
diff --git a/egs2/ms_indic_18/asr1/conf/tuning/decode_asr_transformer.yaml b/egs2/ms_indic_18/asr1/conf/tuning/decode_asr_transformer.yaml
@@ -0,0 +1,7 @@
+batch_size: 1
+beam_size: 10
+penalty: 0.0
+maxlenratio: 0.0
+minlenratio: 0.0
+ctc_weight: 0.4
+lm_weight: 0.4
diff --git a/egs2/ms_indic_18/asr1/conf/tuning/train_asr_transformer.yaml b/egs2/ms_indic_18/asr1/conf/tuning/train_asr_transformer.yaml
@@ -0,0 +1,55 @@
+cudnn_benchmark: false 
+cudnn_deterministic: false 
+use_amp: true 
+# network architecture
+# encoder related
+encoder: transformer
+encoder_conf:
+    input_layer: "conv2d"
+    num_blocks: 6
+    linear_units: 1024
+    dropout_rate: 0.1
+    output_size: 256  # dimension of attention
+    attention_heads: 4
+    attention_dropout_rate: 0.1
+
+# decoder related
+# decoder related
+decoder: transformer
+decoder_conf:
+    input_layer: "embed"
+    num_blocks: 6
+    linear_units: 1024
+    dropout_rate: 0.1
+
+# hybrid CTC/attention
+model_conf:
+    ctc_weight: 0.3
+    lsm_weight: 0.1
+    length_normalized_loss: false
+
+# minibatch related
+batch_type: folded
+batch_size: 256
+
+# optimization related
+optim: adam
+accum_grad: 1
+grad_clip: 5
+patience: 3
+max_epoch: 50
+optim_conf:
+    lr: 0.001
+scheduler: warmuplr
+scheduler_conf:
+    warmup_steps: 25000
+
+# others:
+best_model_criterion:
+-  - valid
+   - acc
+   - max
+keep_nbest_models: 10
+
+init: xavier_uniform # Our empirical studies shows that this initialization
+                     # is very important to low-resource ASR training