зеркало из https://github.com/mozilla/DeepSpeech.git
Update README to include GIF and feature usage of the Python package more prominently
This commit is contained in:
Родитель
1f7b901bad
Коммит
5ce7a2d48f
43
README.md
43
README.md
|
@ -5,14 +5,25 @@
|
|||
|
||||
Project DeepSpeech is an open source Speech-To-Text engine. It uses a model trained by machine learning techniques, based on [Baidu's Deep Speech research paper](https://arxiv.org/abs/1412.5567). Project DeepSpeech uses Google's [TensorFlow](https://www.tensorflow.org/) project to make the implementation easier.
|
||||
|
||||
![Usage](images/usage.gif)
|
||||
|
||||
Pre-built binaries that can be used for performing inference with a trained model can be installed with `pip`. You can then use the `deepspeech` binary to do speech-to-text on an audio file:
|
||||
|
||||
```bash
|
||||
pip install deepspeech
|
||||
deepspeech output_model.pb my_audio_file.wav alphabet.txt
|
||||
```
|
||||
|
||||
See the output of `deepspeech -h` for more information.
|
||||
|
||||
**Table of Contents**
|
||||
|
||||
- [Prerequisites](#prerequisites)
|
||||
- [Getting the code](#getting-the-code)
|
||||
- [Using the model](#using-the-model)
|
||||
- [Using the Python package](#using-the-python-package)
|
||||
- [Using the command line client](#using-the-command-line-client)
|
||||
- [Installing Python bindings](#installing-python-bindings)
|
||||
- [Installing Node.JS bindings](#installing-nodejs-bindings)
|
||||
- [Using the Node.JS package](#using-the-nodejs-package)
|
||||
- [Training](#training)
|
||||
- [Recommendations](#recommendations)
|
||||
- [Training a model](#training-a-model)
|
||||
|
@ -39,6 +50,17 @@ git clone https://github.com/mozilla/DeepSpeech
|
|||
|
||||
If all you want to do is use an already trained model for doing speech-to-text, you can grab one of our pre-built binaries. You can use a command-line binary, a Python package, or a Node.JS package.
|
||||
|
||||
### Using the Python package
|
||||
|
||||
Pre-built binaries that can be used for performing inference with a trained model can be installed with `pip`. You can then use the `deepspeech` binary to do speech-to-text on an audio file:
|
||||
|
||||
```bash
|
||||
pip install deepspeech
|
||||
deepspeech output_model.pb my_audio_file.wav alphabet.txt lm.binary trie
|
||||
```
|
||||
|
||||
See [client.py](native_client/python/client.py) for an example of how to use the package programatically.
|
||||
|
||||
### Using the command-line client
|
||||
|
||||
To download the pre-built binaries, use `util/taskcluster.py`:
|
||||
|
@ -61,22 +83,7 @@ This will download `native_client.tar.xz` which includes the deepspeech binary a
|
|||
|
||||
See the help output with `./deepspeech -h` and the [native client README](native_client/README.md) for more details.
|
||||
|
||||
### Installing Python bindings
|
||||
|
||||
Pre-built binaries that can be used for performing inference with a trained model can be found on TaskCluster. You'll need to download the appropriate Python wheel package.
|
||||
|
||||
[deepspeech-0.0.2-cp27-cp27mu-manylinux1_x86_64.whl (Python 2.7, Linux / amd64)](https://index.taskcluster.net/v1/task/project.deepspeech.deepspeech.native_client.master.cpu/artifacts/public/deepspeech-0.0.2-cp27-cp27mu-manylinux1_x86_64.whl)
|
||||
|
||||
[Other configurations](https://tools.taskcluster.net/index/artifacts/#project.deepspeech.deepspeech.native_client.master/project.deepspeech.deepspeech.native_client.master)
|
||||
|
||||
You can use pip to install the Python package, like so:
|
||||
```bash
|
||||
pip install <path to .whl file>
|
||||
```
|
||||
|
||||
See [client.py](native_client/python/client.py) for an example of how to use the bindings.
|
||||
|
||||
### Installing Node.JS bindings
|
||||
### Using the Node.JS package
|
||||
|
||||
You can download the Node.JS bindings using `util/taskcluster.py` and install them with `npm`:
|
||||
|
||||
|
|
Двоичный файл не отображается.
После Ширина: | Высота: | Размер: 536 KiB |
|
@ -18,13 +18,13 @@ If you're looking to train a model, you now have a `libctc_decoder_with_kenlm.so
|
|||
|
||||
## Installing the language bindings
|
||||
|
||||
`native_client.tar.xz` doesn't include the language bindings by default. For that you can use the `--artifact` parameter to download a specific language binding file.
|
||||
For the Python bindings, you can use `pip`:
|
||||
|
||||
For Python bindings, use `--artifact file_name`, where `file_name` is the appropriate file for your Python version and platform. The names of the available artifacts can be found on the listing page: [Linux](https://tools.taskcluster.net/index/artifacts/project.deepspeech.deepspeech.native_client.master/cpu) or [macOS](https://tools.taskcluster.net/index/artifacts/project.deepspeech.deepspeech.native_client.master/osx).
|
||||
```
|
||||
pip install deepspeech
|
||||
```
|
||||
|
||||
For example, for Python 2.7 bindings on Linux, you can do `python util/taskcluster.py --target /destination --artifact deepspeech-0.0.2-cp27-cp27mu-manylinux1_x86_64.whl`.
|
||||
|
||||
For Node.JS bindings, use `--artifact deepspeech-0.0.2.tgz`.
|
||||
For Node.JS bindings, use `python util/taskcluster.py --target . --artifact deepspeech-0.0.2.tgz` to download the package and `npm install deepspeech-0.0.2.tgz` to install it.
|
||||
|
||||
## Build Requirements
|
||||
|
||||
|
|
Загрузка…
Ссылка в новой задаче