Merge pull request #79 from lissyx/ds0.6.0a15

Ds0.6.0a15

DeepSpeech/Dockerfile.train

@@ -1,8 +1,8 @@
 FROM nvidia/cuda:10.0-cudnn7-runtime-ubuntu18.04
 
 ARG ds_repo=mozilla/DeepSpeech
-ARG ds_branch=336daa1641b8a0bb6c02befa69b2b68b84161338
-ARG ds_sha1=336daa1641b8a0bb6c02befa69b2b68b84161338
+ARG ds_branch=c5935644a1758c5d9e678642eb83106376dc0cd8
+ARG ds_sha1=c5935644a1758c5d9e678642eb83106376dc0cd8
 ARG kenlm_repo=kpu/kenlm
 ARG kenlm_branch=2ad7cb56924cd3c6811c604973f592cb5ef604eb
 
@@ -15,6 +15,7 @@ ARG learning_rate=0.0001
 ARG dropout=0.3
 ARG lm_alpha=0.65
 ARG lm_beta=1.45
+ARG beam_width=500
 ARG early_stop=1
 
 ARG lm_evaluate_range=
@@ -44,6 +45,7 @@ ENV LEARNING_RATE=$learning_rate
 ENV DROPOUT=$dropout
 ENV LM_ALPHA=$lm_alpha
 ENV LM_BETA=$lm_beta
+ENV BEAM_WIDTH=$beam_width
 
 ENV LM_EVALUATE_RANGE=$lm_evaluate_range
 ENV ENGLISH_COMPATIBLE=$english_compatible

DeepSpeech/fr/train.sh

@@ -61,6 +61,22 @@ pushd $HOME/ds/
 			--export_language "fra"
 	fi;
 
+	if [ ! -f "/mnt/models/fr-fr.zip" ]; then
+		python -u DeepSpeech.py \
+			--alphabet_config_path /mnt/models/alphabet.txt \
+			--lm_binary_path /mnt/lm/lm.binary \
+			--lm_trie_path /mnt/lm/trie \
+			--feature_cache /mnt/sources/feature_cache \
+			--n_hidden ${N_HIDDEN} \
+			--beam_width ${BEAM_WIDTH} \
+			--lm_alpha ${LM_ALPHA} \
+			--lm_beta ${LM_BETA} \
+			--checkpoint_dir /mnt/checkpoints/ \
+			--export_dir /mnt/models/fr-fr \
+			--export_zip \
+			--export_language "Français (FR)"
+	fi;
+
 	if [ ! -f "/mnt/models/output_graph.pbmm" ]; then
 		./convert_graphdef_memmapped_format --in_graph=/mnt/models/output_graph.pb --out_graph=/mnt/models/output_graph.pbmm
 	fi;