Test GPT2 AI language translation model

We are developing a new OpenCL-based GPU-powered text to speech model, called Tandem TTS (or GPT2) trained on a large-vocabulary, language modeling task in order to produce more accurate speech output than traditional n-gram models and neural net architectures. We have developed a fast, efficient algorithm for training on OpenCL-based GPUs and demonstrated that realtime decoding can be achieved for short phrases on GPU and CPU devices. This work should help us to create a more robust, more accurate, and more powerful model for generating high quality speech.

Job Description:

You will be tasked with testing a prototype of the Tandem TTS model on a dataset of language samples that we have collected. The model will generate text in one of a few languages depending on the language that is selected. For each sentence, you will have to listen to the generated speech output. The sentence will be followed by the source text, so you will be able to confirm that the model produced accurate translation. You will be able to make a judgement regarding the accuracy of the model using the generated audio. You do NOT need to know the language spoken as it will be written phonetically. You will need to download and install our speech samples, and run them.


After successful test, you will be asked to provide detailed review feedback on the Tandem TTS results for the sample data collected. You will also be asked to provide feedback on possible improvements for the model. This will involve: 1) testing on more language models and 2) modifying the model training technique to improve stability and the overall accuracy of the output.

## Minimal Requirements:

We recommend you to use Windows 10 and above, older platform compatibility not guaranteed.

* A working copy of Windows 10 or above.

* GPU or CPU with the capability to work with OpenCL (all devices are compatible, however the more powerful it is the faster it will finish)

* A sound card with a built-in microphone (the better the microphone quality, the better the results).

* A modern web browser that supports JavaScript.

* Download and install the speech samples.

* Be able to install OpenCL-compatible driver software for your device.

* A working Internet connection.

You will be provided detailed assistance and instructions on the correct installation of the tools and drivers if your device drivers and not installed already.

## Skills & Qualifications:

No pre-requisites. You just need to be able to listen to audio and provide feedback.

## What to Expect:

The tasks you will be required to perform can be done quickly (within minutes) and will require no programming knowledge. You will be asked to perform several tasks that will provide you with feedback on the language model. After completion of the tasks, you will be able to provide feedback on the model’s overall performance and its accuracy. This will provide us with feedback and will help us to make improvements to the algorithm. The total assignments should not take more than an hour.