Next, install the CUDA Toolkit, be sure to install the 10.1 version.
Nvidia gtx gassistant drivers#
Start by installing the NVIDIA GPU drivers, which are just the regular drivers you install with your video card - you may already have these installed, but it's best to double-check. Installing driversĪll of the software we need requires some Nvidia drivers to work, so we will install all of those now. All commands will be run in an Anaconda prompt unless otherwise stated. After installing, you should now be able to open a Conda prompt from the start menu using "Start->Anaconda" (just start typing and it should appear). Start by heading over to the Anaconda downloads page, then download and install the 64-bit version for your platform. To ease the pain, on Windows, we will use Anaconda to make managing the Python environments easier. There are quite a few different software requirements, and some of them are a real pain to get installed correctly as a result of a series of dependency problems. As they are designed with Linux in mind, installation should be much shorter and likely won't have as many gotchas as the Windows procedure. If you do have a Linux machine with a CUDA GPU, most of the instructions will still hold, but a few things will actually be more straightforward. As such, this guide will focus on using Windows as the OS for running both of these programs. However, my regular computer has a GPU that can easily handle these tasks without even blinking. Personally, I have no Linux computer with a CUDA GPU capable of running this stuff. What OS will we use?Īlmost all of the installation instructions for DeepSpeech and Mozilla TTS that can be found online are done with a Linux operating system. Keep in mind that it does not need to be a very powerful GPU in the "gaming" sense as any GPU with CUDA capabilities will provide a significant boost over just using the CPU. You can use this list to see which GPUs will work. Both programs run on TensorFlow which at the time of writing has a required CUDA compute capability of 3.0.
As you can see, a GPU is almost mandatory for this application unless you are willing to accept very long and unnatural delays.Īs such, the hardware requirement is to have a graphics card that supports CUDA (Nvidia only). With a Nvidia GTX 1080, the time to compute a Text to Speech operation was around 2 seconds, compared to 17 seconds for the same sentence using only CPU processing. The key is to use a GPU as it really accelerates things. Keep in mind that's with a pretty strong CPU - especially when compared to the Raspberry Pi!
In my testing using an Intel i7-7700k CPU, both of the required programs ran at a speed that was simply unacceptable for a voice assistant. Unfortunately, both the Speech to Text and the Text to Speech components are very processor intensive and will make quick work of even the best hardware.
0 kommentar(er)
