Nim bindings for the Mozilla's DeepSpeech (v0.7.0) deep learning model.

Prerequisites:

• Shared library libdeepspeech.so for v0.7.0 which you can get after downloading native client package from their releases page based on your cpu architecture and operating system and extracting it.

  Note: Native package would look like native_client.amd64.cpu.win.tar.xz for windows OS on a AMD64 CPU architecture.

  Note: This lib must be in your system's PATH or in the directory from where executable/binary is run from.

• Pretrained Models and Scorer. Download them from their releases page as well which would look like deepspeech-0.7.(0|1).models.pbmm/tflite and deepspeech-0.7.(0|1).scorer

Note:

• .wav files or raw audio data with single channel,sampled at 16Khz and having audio format as INT16 PCM are desired. Be sure to convert or resample data/audio according to above specifications if not already so.

Installation

• $ nimble install deepspeech@0.7.0 or $ nimble install https://gitlab.com/eagledot/nim-deepspeech@0.7.0
• $ nimble install https://gitlab.com/eagledot/nim-wav [optional if you need to read from or write to a .wav file , needed to run examples]

Usage

import deepspeech

For more detailed usage see examples in ./examples/ directory.