BigSpeech

This project is forked from one of the Intel anlalytics-zoo examples.

Thanks for their contributions.

Run deepspeech2 model on BigDL for prediction.

1. Download deepspeech2 model to the 'deepspeech2/model/' directory.

2. Build the project: import the project into IDE or build with "mvn clean package".

3. Start your hdfs service at localhost:9000.

4. Enter the 'deepspeech2' directory.

cd deepspeech2

5. Put the audio file to hdfs

hdfs dfs -mkdir /data
hdfs dfs -put data/1462-170145-0004.flac /data/1462-170145-0004.flac

6. Run 'prediction.sh' to get result

./prediction.sh

Output will looks like:

+------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------+
|path                                            |output                                                                                                                             |
+------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------+
|hdfs://127.0.0.1:9000/data/1462-170145-0004.flac|I SAW THAT INTERVIEW IN THE PAPERYESTERDAY TELLING WHER YOU WERE AND I THOUGHT I HAD TO SEE YOU THAT'S ALL BUT NIGHT I'AM GOING DOW|
+------------------------------------------------+-----------------------------------------------------------------------------------------------------------------------------------+