Taejin Park

Chao-Han Huck Yang, Taejin Park, Yuan Gong et al. · gatech

Given recent advances in generative AI technology, a key question is how large language models (LLMs) can enhance acoustic modeling tasks using text decoding results from a frozen, pretrained automatic speech recognition (ASR) model. To explore new capabilities in language modeling for speech processing, we introduce the generative speech transcription error correction (GenSEC) challenge. This challenge comprises three post-ASR language modeling tasks: (i) post-ASR transcription correction, (ii) speaker tagging, and (iii) emotion recognition. These tasks aim to emulate future LLM-based agents handling voice-based interfaces while remaining accessible to a broad audience by utilizing open pretrained language models or agent-based APIs. We also discuss insights from baseline evaluations, as well as lessons learned for designing future evaluations.

Taejin Park, Ivan Medennikov, Kunal Dhawan et al.

Sortformer is an encoder-based speaker diarization model designed for supervising speaker tagging in speech-to-text models. Instead of relying solely on permutation invariant loss (PIL), Sortformer introduces Sort Loss to resolve the permutation problem, either independently or in tandem with PIL. In addition, we propose a streamlined multi-speaker speech-to-text architecture that leverages Sortformer for speaker supervision, embedding speaker labels into the encoder using sinusoidal kernel functions. This design addresses the speaker permutation problem through sorted objectives, effectively bridging timestamps and tokens to supervise speaker labels in the output transcriptions. Experiments demonstrate that Sort Loss can boost speaker diarization performance, and incorporating the speaker supervision from Sortformer improves multi-speaker transcription accuracy. We anticipate that the proposed Sortformer and multi-speaker architecture will enable the seamless integration of speaker tagging capabilities into foundational speech-to-text systems and multimodal large language models (LLMs), offering an easily adoptable and user-friendly mechanism to enhance their versatility and performance in speaker-aware tasks. The code and trained models are made publicly available through the NVIDIA NeMo Framework.

Nithin Rao Koluguri, Taejin Park, Boris Ginsburg

In this paper, we propose TitaNet, a novel neural network architecture for extracting speaker representations. We employ 1D depth-wise separable convolutions with Squeeze-and-Excitation (SE) layers with global context followed by channel attention based statistics pooling layer to map variable-length utterances to a fixed-length embedding (t-vector). TitaNet is a scalable architecture and achieves state-of-the-art performance on speaker verification task with an equal error rate (EER) of 0.68% on the VoxCeleb1 trial file and also on speaker diarization tasks with diarization error rate (DER) of 1.73% on AMI-MixHeadset, 1.99% on AMI-Lapel and 1.11% on CH109. Furthermore, we investigate various sizes of TitaNet and present a light TitaNet-S model with only 6M parameters that achieve near state-of-the-art results in diarization tasks.

Taejin Park, Kenichi Kumatani, Minhua Wu et al.

Conventional speech enhancement technique such as beamforming has known benefits for far-field speech recognition. Our own work in frequency-domain multi-channel acoustic modeling has shown additional improvements by training a spatial filtering layer jointly within an acoustic model. In this paper, we further develop this idea and use frequency aligned network for robust multi-channel automatic speech recognition (ASR). Unlike an affine layer in the frequency domain, the proposed frequency aligned component prevents one frequency bin influencing other frequency bins. We show that this modification not only reduces the number of parameters in the model but also significantly and improves the ASR performance. We investigate effects of frequency aligned network through ASR experiments on the real-world far-field data where users are interacting with an ASR system in uncontrolled acoustic environments. We show that our multi-channel acoustic model with a frequency aligned network shows up to 18% relative reduction in word error rate.

Taejin Park, Taejin Lee

In this paper, a Blind Source Separation (BSS) algorithm for multichannel audio contents is proposed. Unlike common BSS algorithms targeting stereo audio contents or microphone array signals, our technique is targeted at multichannel audio such as 5.1 and 7.1ch audio. Since most multichannel audio object sources are panned using the Inter-channel Loudness Difference (ILD), we employ the ILVS (Inter-channel Loudness Vector Sum) concept to cluster common signals (such as background music) from each channel. After separating the common signals from each channel, we employ an Expectation Maximization (EM) algorithm with a von-Mises distribution to successfully classify the clustering of sound source objects and separate the audio signals from the original mixture. Our proposed method can therefore separate common audio signals and object source signals from multiple channels with reasonable quality. Our multichannel audio content separation technique can be applied to an upmix system or a cinema audio system requiring multichannel audio source separation.

Taejin Park, Taejin Lee

A new musical instrument classification method using convolutional neural networks (CNNs) is presented in this paper. Unlike the traditional methods, we investigated a scheme for classifying musical instruments using the learned features from CNNs. To create the learned features from CNNs, we not only used a conventional spectrogram image, but also proposed multiresolution recurrence plots (MRPs) that contain the phase information of a raw input signal. Consequently, we fed the characteristic timbre of the particular instrument into a neural network, which cannot be extracted using a phase-blinded representations such as a spectrogram. By combining our proposed MRPs and spectrogram images with a multi-column network, the performance of our proposed classifier system improves over a system that uses only a spectrogram. Furthermore, the proposed classifier also outperforms the baseline result from traditional handcrafted features and classifiers.