nvidia/diar_streaming_sortformer_4spk-v2.1

Streaming Sortformer Diarizer 4spk v2.1

|

This model is a streaming version of Sortformer diarizer. Sortformer[1] is a novel end-to-end neural model for speaker diarization, trained with unconventional objectives compared to existing end-to-end diarization models.

Streaming Sortformer[2] employs an Arrival-Order Speaker Cache (AOSC) to store frame-level acoustic embeddings of previously observed speakers.

Sortformer resolves permutation problem in diarization following the arrival-time order of the speech segments from each speaker.

Discover more from NVIDIA:

For documentation, deployment guides, enterprise-ready APIs, and the latest open models—including Nemotron and other cutting-edge speech, translation, and generative AI—visit the NVIDIA Developer Portal at developer.nvidia.com. Join the community to access tools, support, and resources to accelerate your development with NVIDIA’s NeMo, Riva, NIM, and foundation models.

Explore more from NVIDIA:

What is Nemotron?
NVIDIA Developer Nemotron
NVIDIA Riva Speech
NeMo Documentation

Model Architecture

Streaming sortformer employs pre-encode layer in the Fast-Conformer to generate speaker-cache. At each step, speaker cache is filtered to only retain the high-quality speaker cache vectors.

Aside from speaker-cache management part, streaming Sortformer follows the architecture of the offline version of Sortformer. Sortformer consists of an L-size (17 layers) NeMo Encoder for Speech Tasks (NEST)[3] which is based on Fast-Conformer[4] encoder. Following that, an 18-layer Transformer[5] encoder with hidden size of 192, and two feedforward layers with 4 sigmoid outputs for each frame input at the top layer. More information can be found in the Streaming Sortformer paper[2].

NVIDIA NeMo

To train, fine-tune or perform diarization with Sortformer, you will need to install NVIDIA NeMo[6]. We recommend you install it after you've installed Cython and latest PyTorch version.

apt-get update && apt-get install -y libsndfile1 ffmpeg
pip install Cython packaging
pip install git+https://github.com/NVIDIA/NeMo.git@main#egg=nemo_toolkit[asr]

How to Use this Model

The model is available for use in the NeMo Framework[6], and can be used as a pre-trained checkpoint for inference or for fine-tuning on another dataset.

Loading the Model

from nemo.collections.asr.models import SortformerEncLabelModel

# load model from Hugging Face model card directly (You need a Hugging Face token)
diar_model = SortformerEncLabelModel.from_pretrained("nvidia/diar_streaming_sortformer_4spk-v2")

# If you have a downloaded model in "/path/to/diar_streaming_sortformer_4spk-v2.nemo", load model from a downloaded file
diar_model = SortformerEncLabelModel.restore_from(restore_path="/path/to/diar_streaming_sortformer_4spk-v2.nemo", map_location='cuda', strict=False)

# switch to inference mode
diar_model.eval()

Input Format

Input to Sortformer can be an individual audio file:

audio_input="/path/to/multispeaker_audio1.wav"

or a list of paths to audio files:

audio_input=["/path/to/multispeaker_audio1.wav", "/path/to/multispeaker_audio2.wav"]

or a jsonl manifest file:

audio_input="/path/to/multispeaker_manifest.json"

where each line is a dictionary containing the following fields:

# Example of a line in `multispeaker_manifest.json`
{
    "audio_filepath": "/path/to/multispeaker_audio1.wav",  # path to the input audio file 
    "offset": 0, # offset (start) time of the input audio
    "duration": 600,  # duration of the audio, can be set to `null` if using NeMo main branch
}
{
    "audio_filepath": "/path/to/multispeaker_audio2.wav",  
    "offset": 900,
    "duration": 580,  
}

Setting up Streaming Configuration

Streaming configuration is defined by the following parameters, all measured in 80ms frames:

• CHUNK_SIZE: The number of frames in a processing chunk.
• RIGHT_CONTEXT: The number of future frames attached after the chunk.
• FIFO_SIZE: The number of previous frames attached before the chunk, from the FIFO queue.
• UPDATE_PERIOD: The number of frames extracted from the FIFO queue to update the speaker cache.
• SPEAKER_CACHE_SIZE: The total number of frames in the speaker cache.

Here are recommended configurations for different scenarios:

Configuration	Latency	RTF	CHUNK_SIZE	RIGHT_CONTEXT	FIFO_SIZE	UPDATE_PERIOD	SPEAKER_CACHE_SIZE
very high latency	30.4s	0.002	340	40	40	300	188
low latency	1.04s	0.093	6	7	188	144	188

For clarity on the metrics used in the table:

• Latency: Refers to Input Buffer Latency, calculated as CHUNK_SIZE + RIGHT_CONTEXT. This value does not include computational processing time.
• Real-Time Factor (RTF): Characterizes processing speed, calculated as the time taken to process an audio file divided by its duration. RTF values are measured with a batch size of 1 on an NVIDIA RTX 6000 Ada Generation GPU.

To set streaming configuration, use:

diar_model.sortformer_modules.chunk_len = CHUNK_SIZE
diar_model.sortformer_modules.chunk_right_context = RIGHT_CONTEXT
diar_model.sortformer_modules.fifo_len = FIFO_SIZE
diar_model.sortformer_modules.spkcache_update_period = UPDATE_PERIOD
diar_model.sortformer_modules.spkcache_len = SPEAKER_CACHE_SIZE
diar_model.sortformer_modules._check_streaming_parameters()

Getting Diarization Results

To perform speaker diarization and get a list of speaker-marked speech segments in the format 'begin_seconds, end_seconds, speaker_index', simply use:

predicted_segments = diar_model.diarize(audio=audio_input, batch_size=1)

To obtain tensors of speaker activity probabilities, use:

predicted_segments, predicted_probs = diar_model.diarize(audio=audio_input, batch_size=1, include_tensor_outputs=True)

Input

This model accepts single-channel (mono) audio sampled at 16,000 Hz.

• The actual input tensor is a Ns x 1 matrix for each audio clip, where Ns is the number of samples in the time-series signal.
• For instance, a 10-second audio clip sampled at 16,000 Hz (mono-channel WAV file) will form a 160,000 x 1 matrix.

Output

The output of the model is an T x S matrix, where:

• S is the maximum number of speakers (in this model, S = 4).
• T is the total number of frames, including zero-padding. Each frame corresponds to a segment of 0.08 seconds of audio.
  Each element of the T x S matrix represents the speaker activity probability in the [0, 1] range. For example, a matrix element a(150, 2) = 0.95 indicates a 95% probability of activity for the second speaker during the time range [12.00, 12.08] seconds.

Train and evaluate Sortformer diarizer using NeMo

Training

Sortformer diarizer models are trained on 8 nodes of 8×NVIDIA Tesla V100 GPUs. We use 90 second long training samples and batch size of 4. The model can be trained using this example script and base config.

Inference

Sortformer diarizer models can be performed with post-processing algorithms using inference example script. If you provide the post-processing YAML configs in post_processing folder to reproduce the optimized post-processing algorithm for each development dataset.

Technical Limitations

• The model operates in a streaming mode (online mode).
• It can detect a maximum of 4 speakers; performance degrades on recordings with 5 and more speakers.
• While the model is designed for long-form audio and can handle recordings that are several hours long, performance may degrade on very long recordings.
• The model was trained on publicly available speech datasets, primarily in English. As a result:
  • Performance may degrade on non-English speech.
  • Performance may also degrade on out-of-domain data, such as recordings in noisy conditions.

Datasets

Sortformer was trained on approximately 5,000 hours of audio, combining real conversations and simulated audio mixtures generated using the NeMo speech data simulator[7]. All datasets used in training follow the RTTM labeling format. A subset of the RTTM files were processed specifically for speaker diarization model training. Data collection methods vary across individual datasets. For example, the above datasets include phone calls, interviews, web videos, and audiobook recordings. Please refer to the Linguistic Data Consortium (LDC) website or dataset webpage for detailed data collection methods.

Training Datasets (Real conversations)

• Fisher English (LDC)
• AMI Meeting Corpus (IHM, lapel-mix, SDM) with Forced alignment based ground-truth RTTMs[8]
• VoxConverse-v0.3
• ICSI
• AISHELL-4
• Third DIHARD Challenge Development (LDC)
• 2000 NIST Speaker Recognition Evaluation, split1 (LDC)
• DiPCo
• AliMeeting with Forced alignment based ground-truth RTTMs[8]
• NOTSOFAR1

Training Datasets (Used to simulate audio mixtures)

• 2004-2010 NIST Speaker Recognition Evaluation (LDC)
• Librispeech

Performance

Evaluation data specifications

Dataset	Number of speakers	Number of Sessions
DIHARD III Eval <=4spk	1-4	219
DIHARD III Eval >=5spk	5-9	40
DIHARD III Eval full	1-9	259
CALLHOME-part2 2spk	2	148
CALLHOME-part2 3spk	3	74
CALLHOME-part2 4spk	4	20
CALLHOME-part2 5spk	5	5
CALLHOME-part2 6spk	6	3
CALLHOME-part2 full	2-6	250
CHAES CH109 (2spk set)	2	109
AliMeeting Test	2-4	20
AMI Test	3-4	16
NOTSOFAR1 Eval SC <=4spk	3-4	70
NOTSOFAR1 Eval SC >=5spk	5-7	90
NOTSOFAR1 Eval SC full	3-7	160

Diarization Error Rate (DER)

• All evaluations include overlapping speech.
• Collar tolerance is 0.25s for CALLHOME-part2 and CH109.
• Collar tolerance is 0s for DIHARD III Eval, AliMeeting Test, AMI Test and NOTSOFAR1 Eval.
• Forced alignment based ground-truth RTTMs[8] are used for AMI and AliMeeting.

Evaluation Results (Telephonic and General-Purpose Speech Corpus)

Model	Latency	DIHARD III Eval <=4spk	DIHARD III Eval >=5spk	DIHARD III Eval full	CALLHOME-part2 2spk	CALLHOME-part2 3spk	CALLHOME-part2 4spk	CALLHOME-part2 5spk	CALLHOME-part2 6spk	CALLHOME-part2 full	CH109
diar_streaming_sortformer_4spk-v2	30.4s	14.63	40.74	19.68	6.27	10.27	12.30	19.08	28.09	10.50	5.03
diar_streaming_sortformer_4spk-v2.1	30.4s	14.84	38.90	19.49	5.65	10.03	12.33	22.35	22.26	10.10	5.04
diar_streaming_sortformer_4spk-v2	1.04s	14.49	42.22	19.85	7.51	11.45	13.75	23.22	29.22	11.89	5.37
diar_streaming_sortformer_4spk-v2.1	1.04s	15.09	41.42	20.21	6.65	11.25	13.35	22.12	24.51	11.19	5.09

Evaluation Results (Meeting Speech Corpus)

Model	Latency	AliMeeting Test near	AliMeeting Test far	AMI Test IHM	AMI Test SDM	NOTSOFAR1 Eval SC <=4spk	NOTSOFAR1 Eval SC >=5spk	NOTSOFAR1 Eval full
diar_streaming_sortformer_4spk-v2	30.4s	19.63	21.09	22.39	28.56	23.31	40.49	33.43
diar_streaming_sortformer_4spk-v2.1	30.4s	11.73	13.55	15.90	17.80	15.95	34.81	27.07
diar_streaming_sortformer_4spk-v2	1.04s	19.98	22.09	25.11	31.34	24.41	41.55	34.52
diar_streaming_sortformer_4spk-v2.1	1.04s	12.60	15.60	16.67	20.57	17.26	36.76	28.75

References

[1] Sortformer: Seamless Integration of Speaker Diarization and ASR by Bridging Timestamps and Tokens

[2] Streaming Sortformer: Speaker Cache-Based Online Speaker Diarization with Arrival-Time Ordering

[3] NEST: Self-supervised Fast Conformer as All-purpose Seasoning to Speech Processing Tasks

[4] Fast Conformer with Linearly Scalable Attention for Efficient Speech Recognition

[5] NVIDIA NeMo Framework

[6] NeMo speech data simulator

[7] Can We Really Repurpose Multi-Speaker ASR Corpus for Speaker Diarization?

Licence

Use of this model is governed by the NVIDIA Open Model License Agreement.