VoxLingua107Identifier<T>

Models & Types · AiDotNet.Audio.LanguageIdentification

VoxLingua107 language identifier supporting 107 languages.

Inheritance: NeuralNetworkBase<T> → AudioNeuralNetworkBase<T> → VoxLingua107Identifier<T>

Implements: ILanguageIdentifier<T>

For Beginners

VoxLingua107 is like having a polyglot friend who can recognize 107 different languages just by listening.

Key features:

  • Covers most of the world’s major languages
  • Trained on real-world YouTube audio (diverse accents and recording conditions)
  • Can identify languages even from short clips (3-10 seconds)
  • Handles code-switching and multilingual speakers

Example usage:

How It Works

VoxLingua107 is a language identification model trained on the VoxLingua107 dataset, which contains speech samples from 107 languages collected from YouTube videos. The model uses the ECAPA-TDNN architecture and is specifically optimized for large-scale multilingual language identification.

Supported language families include:

  • Indo-European (English, Spanish, French, German, Russian, Hindi, etc.)
  • Sino-Tibetan (Mandarin, Cantonese, etc.)
  • Afro-Asiatic (Arabic, Hebrew, Amharic, etc.)
  • Austronesian (Indonesian, Tagalog, Malay, etc.)
  • Niger-Congo (Swahili, Yoruba, Zulu, etc.)
  • Altaic (Turkish, Korean, Japanese, Mongolian, etc.)
  • And many more…

Constructors

Constructor Summary
VoxLingua107Identifier(NeuralNetworkArchitecture<T> architecture, VoxLingua107Options options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) Creates a VoxLingua107 identifier for native training.
VoxLingua107Identifier(NeuralNetworkArchitecture<T> architecture, string modelPath, VoxLingua107Options options) Creates a VoxLingua107 identifier with ONNX model for inference.

Properties

Property Summary
IReadOnlyList<string> SupportedLanguages { get; }
bool SupportsTraining { get; }
int EmbeddingDimension { get; } Gets the embedding dimension.
int NumLanguages { get; } Gets the number of languages the classifier head predicts.

Methods

Method Summary
IReadOnlyDictionary<string, T> GetLanguageProbabilities(Tensor<T> audio)
IReadOnlyList<LanguageSegment<T>> IdentifyLanguageSegments(Tensor<T> audio, int windowSizeMs)
IReadOnlyList<ValueTuple<string, T>> GetTopLanguages(Tensor<T> audio, int topN)
IReadOnlyList<string> GetLanguagesByFamily(string family) Gets all languages in a specific language family.
LanguageResult<T> IdentifyLanguage(Tensor<T> audio)
ModelMetadata<T> GetModelMetadata()
Tensor<T> ForwardForTraining(Tensor<T> input)
ValueTuple<bool, T> AreSameLanguage(Tensor<T> audio1, Tensor<T> audio2)
string GetLanguageDisplayName(string languageCode)

Fields

Field Summary
string[] VoxLingua107Languages The 107 language codes supported by VoxLingua107 (ISO 639-1/3).