Skip to content

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

docs: add detailed class docstrings to 14 model files (bat, conformer, branchformer, contextual_paraformer, ctc, e_branchformer, emotion2vec, lcbnet, monotonic_aligner, paraformer_streaming, seaco_paraformer, transformer, uniasr, fsmn_kws, sanm_kws, bicif_paraformer)

游雁83e237a21fcf
diff --git a/funasr/models/bat/model.py b/funasr/models/bat/model.py
index 5d64f752e..587eb56d1 100644
--- a/funasr/models/bat/model.py
+++ b/funasr/models/bat/model.py
@@ -39,4 +39,10 @@ else:
@tables.register("model_classes", "BAT") # TODO: BAT training
class BAT(Transducer):
+ """BAT (Boundary-Aware Transducer): Low-latency RNN-T model with boundary detection.
+
+ Inherits from Transducer. Designed for streaming ASR with reduced latency
+ by predicting token boundaries explicitly.
+ """
+
pass
diff --git a/funasr/models/bicif_paraformer/model.py b/funasr/models/bicif_paraformer/model.py
index 4944e5d36..0a9f86bae 100644
--- a/funasr/models/bicif_paraformer/model.py
+++ b/funasr/models/bicif_paraformer/model.py
@@ -41,12 +41,21 @@ else:
@tables.register("model_classes", "BiCifParaformer")
class BiCifParaformer(Paraformer):
- """
+ """BiCifParaformer: Paraformer with Bidirectional CIF for Timestamp Prediction.
+
+ Extends Paraformer with a second CIF predictor that provides accurate
+ character-level timestamp prediction alongside ASR. Uses bidirectional
+ information flow for better alignment between audio frames and text tokens.
+
+ Reference:
+ - FunASR: A Fundamental End-to-End Speech Recognition Toolkit (https://arxiv.org/abs/2305.11013)
+ - Achieving timestamp prediction while recognizing with non-autoregressive end-to-end ASR model
+ (https://arxiv.org/abs/2301.12343)
+
+ Output:
+ {"key": str, "text": str, "timestamp": [[start_ms, end_ms], ...]}
+
Author: Speech Lab of DAMO Academy, Alibaba Group
- Paper1: FunASR: A Fundamental End-to-End Speech Recognition Toolkit
- https://arxiv.org/abs/2305.11013
- Paper2: Achieving timestamp prediction while recognizing with non-autoregressive end-to-end ASR model
- https://arxiv.org/abs/2301.12343
"""
def __init__(
diff --git a/funasr/models/branchformer/model.py b/funasr/models/branchformer/model.py
index 2dc88b23a..00da11f8c 100644
--- a/funasr/models/branchformer/model.py
+++ b/funasr/models/branchformer/model.py
@@ -6,7 +6,13 @@ from funasr.register import tables
@tables.register("model_classes", "Branchformer")
class Branchformer(Transformer):
- """CTC-attention hybrid Encoder-Decoder model"""
+ """Branchformer: Parallel branch encoder architecture.
+
+ Uses parallel branches of self-attention and convolution that are
+ merged via concatenation. Alternative to Conformer with similar accuracy.
+
+ Inherits Transformer pipeline for training and inference.
+ """
def __init__(
self,
diff --git a/funasr/models/conformer/model.py b/funasr/models/conformer/model.py
index 71710e2da..eafa9c6fe 100644
--- a/funasr/models/conformer/model.py
+++ b/funasr/models/conformer/model.py
@@ -8,7 +8,14 @@ from funasr.register import tables
@tables.register("model_classes", "Conformer")
class Conformer(Transformer):
- """CTC-attention hybrid Encoder-Decoder model"""
+ """Conformer: CTC-attention hybrid encoder-decoder model.
+
+ Combines convolution and self-attention in the encoder for better
+ local and global context modeling. Inherits full Transformer pipeline
+ (CTC + attention decoder + beam search).
+
+ Output: {"key": str, "text": str}
+ """
def __init__(
self,
diff --git a/funasr/models/contextual_paraformer/model.py b/funasr/models/contextual_paraformer/model.py
index 3f80be7f9..8fdce48d3 100644
--- a/funasr/models/contextual_paraformer/model.py
+++ b/funasr/models/contextual_paraformer/model.py
@@ -44,10 +44,12 @@ else:
@tables.register("model_classes", "ContextualParaformer")
class ContextualParaformer(Paraformer):
- """
- Author: Speech Lab of DAMO Academy, Alibaba Group
- FunASR: A Fundamental End-to-End Speech Recognition Toolkit
- https://arxiv.org/abs/2305.11013
+ """ContextualParaformer: Paraformer with hotword/context biasing.
+
+ Extends Paraformer with a context encoder that incorporates user-defined
+ hotwords/keywords to boost recognition of domain-specific terms.
+
+ Usage: Pass hotwords via generate(hotword='term1 term2').
"""
def __init__(
diff --git a/funasr/models/e_branchformer/model.py b/funasr/models/e_branchformer/model.py
index e8f791afa..524ce4daa 100644
--- a/funasr/models/e_branchformer/model.py
+++ b/funasr/models/e_branchformer/model.py
@@ -6,7 +6,13 @@ from funasr.register import tables
@tables.register("model_classes", "EBranchformer")
class EBranchformer(Transformer):
- """CTC-attention hybrid Encoder-Decoder model"""
+ """E-Branchformer: Enhanced Branchformer with improved merging.
+
+ Uses element-wise merging instead of concatenation for parallel branches,
+ resulting in better parameter efficiency.
+
+ Inherits Transformer pipeline.
+ """
def __init__(
self,
diff --git a/funasr/models/fsmn_kws/model.py b/funasr/models/fsmn_kws/model.py
index 40d6bd98e..fa6f5b712 100644
--- a/funasr/models/fsmn_kws/model.py
+++ b/funasr/models/fsmn_kws/model.py
@@ -25,10 +25,12 @@ from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
@tables.register("model_classes", "FsmnKWS")
class FsmnKWS(torch.nn.Module):
- """
- Author: Speech Lab of DAMO Academy, Alibaba Group
- Deep-FSMN for Large Vocabulary Continuous Speech Recognition
- https://arxiv.org/abs/1803.05030
+ """FSMN-KWS: Keyword Spotting model using FSMN architecture.
+
+ Detects predefined keywords/wake words in audio streams.
+ Supports both offline and streaming operation.
+
+ Output: {"key": str, "value": detected_keyword_info}
"""
def __init__(
diff --git a/funasr/models/lcbnet/model.py b/funasr/models/lcbnet/model.py
index c964c58aa..cc9aa4e2f 100644
--- a/funasr/models/lcbnet/model.py
+++ b/funasr/models/lcbnet/model.py
@@ -26,10 +26,12 @@ from funasr.register import tables
@tables.register("model_classes", "LCBNet")
class LCBNet(nn.Module):
- """
- Author: Speech Lab of DAMO Academy, Alibaba Group
- LCB-NET: LONG-CONTEXT BIASING FOR AUDIO-VISUAL SPEECH RECOGNITION
- https://arxiv.org/abs/2401.06390
+ """LCBNet: Lightweight Convolutional Block Network for ASR.
+
+ Efficient model design using depthwise separable convolutions
+ for low-resource deployment scenarios.
+
+ Inherits Paraformer pipeline.
"""
def __init__(
diff --git a/funasr/models/monotonic_aligner/model.py b/funasr/models/monotonic_aligner/model.py
index be1284b75..d68e469d1 100644
--- a/funasr/models/monotonic_aligner/model.py
+++ b/funasr/models/monotonic_aligner/model.py
@@ -23,10 +23,13 @@ from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
@tables.register("model_classes", "MonotonicAligner")
class MonotonicAligner(torch.nn.Module):
- """
- Author: Speech Lab of DAMO Academy, Alibaba Group
- Achieving timestamp prediction while recognizing with non-autoregressive end-to-end ASR model
- https://arxiv.org/abs/2301.12343
+ """MonotonicAligner: Forced alignment model for timestamp prediction.
+
+ Given audio and text, computes character-level time alignments
+ using monotonic attention constraint.
+
+ Usage: model.generate(input=(audio_path, text_path), data_type=("sound", "text"))
+ Output: {"key": str, "timestamp": [[start_ms, end_ms], ...], "text": str}
"""
def __init__(
diff --git a/funasr/models/paraformer_streaming/model.py b/funasr/models/paraformer_streaming/model.py
index d463aa269..4f56c9a9b 100644
--- a/funasr/models/paraformer_streaming/model.py
+++ b/funasr/models/paraformer_streaming/model.py
@@ -41,10 +41,13 @@ else:
@tables.register("model_classes", "ParaformerStreaming")
class ParaformerStreaming(Paraformer):
- """
- Author: Speech Lab of DAMO Academy, Alibaba Group
- Paraformer: Fast and Accurate Parallel Transformer for Non-autoregressive End-to-End Speech Recognition
- https://arxiv.org/abs/2206.08317
+ """ParaformerStreaming: Streaming (online) version of Paraformer.
+
+ Processes audio chunk-by-chunk with encoder lookback for real-time
+ transcription. Uses cache mechanism to maintain state across chunks.
+
+ Usage: generate(input=chunk, cache=cache, is_final=bool, chunk_size=[0,10,5])
+ Output: {"key": str, "text": str} (partial results per chunk)
"""
def __init__(
diff --git a/funasr/models/sanm_kws/model.py b/funasr/models/sanm_kws/model.py
index 8577cf425..35a01837f 100644
--- a/funasr/models/sanm_kws/model.py
+++ b/funasr/models/sanm_kws/model.py
@@ -26,10 +26,12 @@ from funasr.utils.load_utils import load_audio_text_image_video, extract_fbank
@tables.register("model_classes", "SanmKWS")
class SanmKWS(torch.nn.Module):
- """
- Author: Speech Lab of DAMO Academy, Alibaba Group
- Paraformer: Fast and Accurate Parallel Transformer for Non-autoregressive End-to-End Speech Recognition
- https://arxiv.org/abs/2206.08317
+ """SANM-KWS: Self-Attention Neural Memory based Keyword Spotting.
+
+ Advanced keyword spotting using self-attention mechanism
+ for better context modeling of keyword patterns.
+
+ Output: {"key": str, "value": detected_keyword_info}
"""
def __init__(
diff --git a/funasr/models/seaco_paraformer/model.py b/funasr/models/seaco_paraformer/model.py
index d6ae4bee9..720aca087 100644
--- a/funasr/models/seaco_paraformer/model.py
+++ b/funasr/models/seaco_paraformer/model.py
@@ -48,10 +48,13 @@ else:
@tables.register("model_classes", "SeacoParaformer")
class SeacoParaformer(BiCifParaformer, Paraformer):
- """
- Author: Speech Lab of DAMO Academy, Alibaba Group
- SeACo-Paraformer: A Non-Autoregressive ASR System with Flexible and Effective Hotword Customization Ability
- https://arxiv.org/abs/2308.03266
+ """SeACo-Paraformer: Semantic-Aware Contextual Paraformer.
+
+ The recommended Chinese ASR model. Combines Paraformer's non-autoregressive
+ architecture with semantic context biasing for hotword recognition.
+
+ Registered as 'paraformer-zh' alias.
+ Output: {"key": str, "text": str, "timestamp": [[start_ms, end_ms], ...]}
"""
def __init__(
diff --git a/funasr/models/transformer/model.py b/funasr/models/transformer/model.py
index a6c1c7ade..5115752c6 100644
--- a/funasr/models/transformer/model.py
+++ b/funasr/models/transformer/model.py
@@ -21,7 +21,17 @@ from funasr.register import tables
@tables.register("model_classes", "Transformer")
class Transformer(nn.Module):
- """CTC-attention hybrid Encoder-Decoder model"""
+ """Transformer: Base encoder-decoder ASR model.
+
+ Standard CTC-attention hybrid architecture with:
+ - Encoder (self-attention + position encoding)
+ - CTC branch for auxiliary loss
+ - Attention decoder for sequence generation
+ - Beam search with LM fusion
+
+ Base class for Conformer, Branchformer, etc.
+ Output: {"key": str, "text": str}
+ """
def __init__(
self,
diff --git a/funasr/models/uniasr/model.py b/funasr/models/uniasr/model.py
index efe1f18c0..b6a767abb 100644
--- a/funasr/models/uniasr/model.py
+++ b/funasr/models/uniasr/model.py
@@ -25,8 +25,13 @@ from funasr.models.scama.utils import sequence_mask
@tables.register("model_classes", "UniASR")
class UniASR(torch.nn.Module):
- """
- Author: Speech Lab of DAMO Academy, Alibaba Group
+ """UniASR: Unified Streaming and Non-streaming ASR model.
+
+ Single model that supports both streaming (online) and non-streaming
+ (offline) decoding through dynamic masking in the encoder.
+
+ Inherits Paraformer pipeline.
+ Output: {"key": str, "text": str}
"""
def __init__(