OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
Audio & Voice AI

DeepSpeech’s documentation

DeepSpeech is an open-source speech-to-text engine built with machine-learning techniques and TensorFlow. Its documentation covers pretrained models, inference through programming-language bindings and command-line tools, and training models from prepared speech data.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

DeepSpeech’s documentation — OpenViST project profile in Audio & Voice AI.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use DeepSpeech when you need a documented engine and client options for local speech-to-text inference, or when you are preparing speech data for model training. Check model/runtime compatibility and whether the training data fits the intended speakers and vocabulary.

Who it is for

  • Developers adding speech transcription to Python, Node.js or Electron applications.
  • Users who need command-line transcription of audio with a pretrained model.
  • Teams preparing speech datasets to train or adapt a speech-recognition model.

Documented uses

  • Convert an audio file into text using a pretrained model.
  • Integrate speech-to-text inference into a Python or Node.js application.
  • Prepare Common Voice data into training, validation and test manifests.
  • Train or adapt a speech model and create an external scorer for a constrained vocabulary.
Before you choose

Requirements & limits

  • The pretrained English model and external scorer are separate files; the scorer is optional during inference.
  • The .pbmm and .tflite model formats require compatible TensorFlow or TensorFlow Lite clients.
  • Python and Node.js clients use SoX for 16 kHz resampling; other runtime dependencies may also be required.
  • GPU inference builds list CUDA 10.1 and CuDNN 7.6 prerequisites, while training guidance lists CUDA 10.0 and CuDNN 7.6.
  • Training-data speaker demographics and transcript text should suit the intended use; fine-tuning or a tailored scorer may be needed when they do not.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Transcribe an audio file with the Python package

Install the Python binding, then run the documented command-line client with a pretrained model and audio file.

Before you start

  • Python 3.5 or later virtual environment is recommended.
  • Install the DeepSpeech Python package with pip3.
  • Download the pretrained English model and scorer files.
  • SoX is used by the Python client to resample audio to 16 kHz; runtime dependencies may include system libraries.

Steps

  1. Create and activate a Python virtual environment.

  2. Install the package with pip3 install deepspeech.

  3. Download the model and scorer files from the DeepSpeech releases, or use the documented wget commands.

  4. Run deepspeech with the model, scorer and audio file arguments; the scorer argument is optional.

Expected outcome

The command-line client performs speech-to-text inference on the specified audio file using the selected model.

Guide limitations

  • The documented command assumes the model files have been downloaded.
  • Use a model compatible with the installed client; model versions can be incompatible after breaking changes.
  • A scorer is optional, and the cited workflow documents a pretrained English model.
Documentation 1 ↗
Source-based practical guide

Prepare Common Voice data for DeepSpeech training

Convert an extracted Common Voice v2.0 language dataset into the CSV splits DeepSpeech training expects.

Before you start

  • Python 3.6 and a Mac or Linux environment.
  • DeepSpeech training code and its dependencies installed as described in the training documentation.
  • An extracted Common Voice v2.0 dataset containing TSV files and referenced audio clips.
  • A filter alphabet may be supplied; it is optional.

Steps

  1. Download and extract a Common Voice v2.0 language dataset.

  2. Run bin/import_cv2.py on the extracted data, optionally providing a filter alphabet.

  3. Use the generated clips/train.csv, clips/dev.csv and clips/test.csv paths as the training, development and test files.

Expected outcome

The importer produces WAV audio files and CSV manifests suitable for passing to DeepSpeech training parameters.

Guide limitations

  • The documented importer is for Common Voice v2.0.
  • Dataset downloading and preprocessing can take a long time, and corpora can be large.
  • The default label-validation function is intended for English; other languages may need locale-specific handling.
Documentation 1 ↗
COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

ProjectDocumented purposePlatformAccess and requirements
DeepSpeech’s documentationConvert an audio file into text using a pretrained model.CLI, Linux, macOS, Windows
EdgedictDecode speech from a YouTube Live video into text.Cloud
Evidence and transparency

Sources & review

deepspeech.readthedocs.io ↗Guide documentation ↗Guide documentation ↗
Source record details
Source availability
Active
Sources checked
12 October 2026
Original website title
Welcome to DeepSpeech’s documentation! — Mozilla DeepSpeech 0.9.3 documentation · Official homepage title · retained in its source language
How we review sourcesSuggest a correction