AI-Studio/app/MindWork AI Studio/Models/Kinds/TranscriptionModelsFamily.cs

using AIStudio.Provider;

namespace AIStudio.Models.Kinds;

/// <summary>
/// The models which listen and write down what they heard.
/// </summary>
/// <remarks>
/// Whisper and Voxtral are missing here on purpose: both have a family of their own, where the
/// statement that they transcribe stands next to what they can do. Repeating it here would be a
/// second place to keep it right.
/// </remarks>
public sealed class TranscriptionModelsFamily : ModelFamily
{
    /// <inheritdoc />
    public override ModelVendor Vendor => ModelVendor.UNKNOWN;

    /// <inheritdoc />
    public override ModelSource Source => new("https://huggingface.co/models?pipeline_tag=automatic-speech-recognition", new DateOnly(2026, 9, 19), "Ported from the transcription markers of Provider/ModelKindExtensions.cs, minus the two which their own families now state. Canary and asr came later: the markers named neither, so both reached the answer meant for everything nobody wrote a rule for. Alibaba's speech line is documented at https://www.alibabacloud.com/help/en/model-studio/qwen-asr-api-reference.");

    /// <inheritdoc />
    protected override void Declare(ModelFamilyBuilder builder)
    {
        // OpenAI appends it to the model it grew out of: gpt-4o-transcribe, gpt-4o-mini-transcribe.
        builder.Modifier("transcribe").AsSegment().Kind(ModelKind.TRANSCRIPTION);

        builder.Modifier("wav2vec").AsSubstring().Inherits();

        builder.Modifier("parakeet").AsSubstring().Inherits();

        // NVIDIA's other line of speech models, written plain: canary-1b, canary-1b-flash,
        // canary-180m-flash. A segment rather than a substring, because canary is an ordinary
        // English word which would otherwise reach into names it has nothing to do with -- the
        // same reason the embedding family gives for gte.
        builder.Modifier("canary").AsSegment().Inherits();

        //
        // The abbreviation the whole field goes by, and the one Alibaba names its speech line
        // after: qwen3-asr-flash, qwen3-asr-1.7b, fun-asr-realtime. Three letters, so a name part
        // and never a substring -- "laser" and "eraser" carry them without meaning any of this.
        //
        builder.Modifier("asr").AsSegment().Inherits();

        //
        // Alibaba also builds the line into its audio models, and "audio" is the longer word, so
        // without this the speech synthesis rule would answer for a model which only listens. A
        // name carrying both words is a transcription model whatever else it is called.
        //
        builder.Modifier("audio").AsSegment().AlsoContains("asr").Inherits();
    }
}