{"id":6932,"date":"2019-12-22T11:04:20","date_gmt":"2019-12-22T11:04:20","guid":{"rendered":"https:\/\/gtechbooster.com\/?p=6932"},"modified":"2023-04-01T01:36:47","modified_gmt":"2023-04-01T01:36:47","slug":"deepspeech-gets-smaller","status":"publish","type":"post","link":"https:\/\/gtechbooster.com\/deepspeech-gets-smaller\/","title":{"rendered":"DeepSpeech Gets Smaller"},"content":{"rendered":"\n

Mozilla DeepSpeech has been updated with support for TensorFlow Lite,\n resulting in a smaller package size and faster performance on some \nplatforms.<\/p>\n\n\n\n

\"\"<\/a><\/div><\/div>

DeepSpeech is a deep leaning-based automatic speech recognition (ASR) engine with a simple API developed by Mozilla. The speech recognition technology and trained models in DeepSpeech are openly available to developers. Mozilla also provides pre-trained English models.<\/p>\n\n\n\n

The latest release, version v0.6, comes with support for TensorFlow \nLite, the version of TensorFlow that\u2019s optimized for mobile and embedded\n devices. This has reduced the DeepSpeech package size from 98 MB to 3.7\n MB, and cut the English model size from 188 MB to 47 MB. The developers\n achieved the cut using post-training quantization, a technique to \ncompress model weights after training is done.<\/p>\n\n\n\n

While TensorFlow Lite is designed for mobile and embedded devices, \nthe DeepSpeech team found it also made DeepSpeech faster on desktop \nplatforms, so they’ve made it available on Windows, macOS, and Linux as \nwell as Raspberry Pi and Android. DeepSpeech v0.6 with TensorFlow Lite \nruns faster than real time on a single core of a Raspberry Pi 4. It also\n uses 22 times less memory.<\/p>\n\n\n\n

The move to TensorFlow 1.14 means the developers have been able to \nmake use of the CuDNN RNN APIs for their training code. This change has \ngiven them around two times faster training times, which means faster \nexperimentation and better models. Support for online feature \naugmentation has also been added.<\/p>\n\n\n\n

Along with the performance improvements, the new decoder exposes \ntiming and confidence metadata, providing new possibilities for \napplications. The new release also includes an extended set of functions\n in the API, not just the textual transcript. You also get metadata \ntiming information for each character in the transcript, and a \nper-sentence confidence value.<\/p>\n\n\n\n

DeepSpeech v0.6 now offers packages for Windows, with .NET, Python, JavaScript, and C bindings. The .NET package is available in the NuGet Gallery, and you can install it directly from Visual Studio.<\/p>\n\n\n\n

More Information<\/h2>\n\n\n\n