Steps to run

Download ds2 model :

nervna model: https://drive.google.com/drive/folders/0B9zID9CU9HQeM1M1SXpmN3oyLUU

download it and unzip into the data folder

build jar

In repo main directory, run mvn clean package and get deepspeech2-bigdl-0.7-SNAPSHOT-jar-with-dependencies.jar

Run Nervanainference with example:

set SPARK_HOME(version recommended: 2.0.1) and put deepspeech2-bigdl-0.7-SNAPSHOT-jar-with-dependencies.jar in current repo main directory

run the following script for nervana ds2 inference with bigdl:

 $SPARK_HOME/bin/spark-submit --master local[1] \
   --conf spark.driver.memory=15g \
   --conf "spark.serializer=org.apache.spark.serializer.JavaSerializer" \
   --driver-class-path deepspeech2-bigdl-0.7-SNAPSHOT-jar-with-dependencies.jar \
   --class com.intel.analytics.zoo.pipeline.deepspeech2.example.NervanaInferenceExample \
   deepspeech2-bigdl-0.7-SNAPSHOT-jar-with-dependencies.jar  \
   -m data/ \
   -d data/1462-170145-0004.flac -n 1 -p 1 -s 30

Performance on dev-clean dataset:

• time: 1201.823576362
• cer: 0.1646082949308756
• wer: 0.41586538461538464