{"id":234083,"date":"2025-05-10T05:38:16","date_gmt":"2025-05-10T05:38:16","guid":{"rendered":"https:\/\/peraltafinancing.com\/analytics\/dia-1-6b-tts-best-text-to-dialogue-generation-model\/"},"modified":"2025-05-10T05:38:16","modified_gmt":"2025-05-10T05:38:16","slug":"dia-1-6b-tts-best-text-to-dialogue-generation-model","status":"publish","type":"post","link":"https:\/\/fivemor.com\/?p=234083","title":{"rendered":"Dia-1.6B TTS : Best Text-to-Dialogue Generation Model"},"content":{"rendered":"


\n<\/p>\n

\n

Looking for the right text-to-speech model<\/a>? The 1.6 billion parameter model Dia might be the one for you. You\u2019d also be surprised to hear that this model was created by two undergraduates and with zero funding! In this article, you\u2019ll learn about the model, how to access and use the model and also see the results to really know what this model is capable of. Before using the model, it would be appropriate to get acquainted with it.<\/p>\n

What is Dia-1.6B?<\/h2>\n

The models trained with the goal of having text as input and natural speech as output, are called text-to-speech models. The Dia-1.6B parameter model developed by Nari Labs belongs to the text-to-speech models family. This is an interesting model that is capable of generating realistic dialogue from a transcript. It\u2019s also worth noting that the model can produce nonverbal communications like laugh, sneeze, whistle etc. Exciting isn\u2019t it?\u00a0<\/p>\n

How to Access the Dia-1.6B?<\/h2>\n

Two ways in which we can access the Dia-1.6B model:<\/p>\n

    \n
  1. Using Hugging Face API with Google Collab<\/a><\/li>\n
  2. Using Hugging Face Spaces<\/li>\n<\/ol>\n

    The first one would require getting the API key and then integrating it in Google Collab with code. The latter is a no-code and allows us to interactively use Dia-1.6B.\u00a0<\/p>\n

    1. Using Hugging Face and Collab<\/h3>\n

    The model is available on Hugging Face and can be run with the help of 10 GB of VRAM, provided by the T4 GPU in Google Collab notebook. We\u2019ll demonstrate the same with a mini conversation.<\/p>\n

    Before we begin, let\u2019s get our Hugging Face access token which will be required to run the code. Go to https:\/\/huggingface.co\/settings\/tokens<\/a> and generate a key, if you don\u2019t have one already.\u00a0<\/p>\n

    Make sure to enable the following permissions:<\/p>\n

    \n
    \"Enabling<\/figure>\n<\/div>\n

    Open a new notebook in Google Collab and add this key in the secrets (Name should be HF_Token):<\/p>\n

    \n
    \"Adding<\/figure>\n<\/div>\n

    Note:<\/strong> Switch to T4 GPU to run this notebook. Then only you\u2019d be able to use the 10GB of VRAM, required for running this model.\u00a0<\/p>\n

    Let\u2019s now get our hands on the the model:<\/p>\n

      \n
    1. First clone the Dia\u2019s Git repository:<\/li>\n<\/ol>\n
      !git clone https:\/\/github.com\/nari-labs\/dia.git<\/code><\/pre>\n
        \n
      1. Install the local package:<\/li>\n<\/ol>\n
        !pip install .\/dia<\/code><\/pre>\n
          \n
        1. Install the soundfile audio library:<\/li>\n<\/ol>\n
          !pip install soundfile<\/code><\/pre>\n
          After running the previous commands, restart the session before proceeding.<\/p>\n
            \n
          1. After the installations, let\u2019s do the necessary imports and initialize the model:<\/li>\n<\/ol>\n
            import soundfile as sf\n\nfrom dia.model import Dia\n\nimport IPython.display as ipd\n\nmodel = Dia.from_pretrained(\"nari-labs\/Dia-1.6B\")<\/code><\/pre>\n
              \n
            1. Initialize the text for the text to speech conversion:<\/li>\n<\/ol>\n
              text = \"[S1] This is how Dia sounds. (laugh) [S2] Don't laugh too much. [S1] (clears throat) Do share your thoughts on the model.\"<\/code><\/pre>\n
                \n
              1. Run inference on the model:<\/li>\n<\/ol>\n
                output = model.generate(text)\n\nsampling_rate = 44100 # Dia uses 44.1Khz sampling rate.\n\noutput_file=\"dia_sample.mp3\"\n\nsf.write(output_file, output, sampling_rate) # Saving the audio\n\nipd.Audio(output_file) # Displaying the audio<\/code><\/pre>\n
                Output:<\/strong><\/p>\n