Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

1.1 KB
Raw Blame
# network architecture
model: Transducer
model_conf:
    auxiliary_ctc_weight: 0.0

# encoder
encoder: RWKVEncoder
encoder_conf:
    kernel: 3
    subsampling_factor: 4
    output_size: 512
    num_blocks: 18
    time_reduction_factor: 2
    att_dropout_rate: 0.1
    ffn_dropout_rate: 0.1
    dropout_rate: 0.1

# decoder (prediction network)
decoder: rnnt_decoder
decoder_conf:
    embed_size: 512
    hidden_size: 512
    embed_dropout_rate: 0.1
    dropout_rate: 0.1
    use_embed_mask: false

# joint network
joint_network: joint_network
joint_network_conf:
    joint_space_size: 512

frontend: WavFrontend
frontend_conf:
    fs: 16000
    window: hamming
    n_mels: 80
    frame_length: 25
    frame_shift: 10
    lfr_m: 1
    lfr_n: 1
    upsacle_samples: true

specaug: SpecAugLFR
specaug_conf:
    apply_time_warp: false
    time_warp_window: 5
    time_warp_mode: bicubic
    apply_freq_mask: true
    freq_mask_width_range:
    - 0
    - 30
    lfr_rate: 6
    num_freq_mask: 1
    apply_time_mask: true
    time_mask_width_range:
    - 0
    - 12
    num_time_mask: 1

tokenizer: CharTokenizer
tokenizer_conf:
  unk_symbol: <unk>
  split_with_space: true