---
title: 'GenTSE: Hierarchical Generative TSE Model'
url: https://www.emergentmind.com/topics/gentse
type: topic
---

# GenTSE: Hierarchical Generative TSE Model

GenTSE is a two-stage, decoder-only, generative language model specifically designed for target speaker extraction (TSE) from speech mixtures. It introduces a hierarchical coarse-to-fine strategy in which semantic and acoustic structures are explicitly disentangled, leveraging continuous speech self-supervised learning (SSL) and codec embeddings for improved generation stability and perceptual quality. GenTSE integrates novel training strategies—Frozen-LM Conditioning (FLC) and Direct Preference Optimization (DPO)—to address exposure bias and align model outputs with human perceptual preferences, respectively. Empirical results on Libri2Mix demonstrate state-of-the-art performance relative to previous LM-based and discriminative TSE baselines, setting new benchmarks for quality, intelligibility, and speaker consistency [2512.20978].

## 1. Hierarchical Architecture and Joint Generation Formulation

GenTSE factorizes the conditional target speech generation process into two sequential, autoregressive, decoder-only transformer stages: a **semantic stage** producing discrete semantic tokens, and an **acoustic stage** generating fine-resolution codec tokens. The model is conditioned on both reference and mixture speech inputs, both represented via rich, frame-level continuous embeddings.

Given $X=(x_{\rm ref},x_{\rm mix})$ (reference and mixture waveforms), GenTSE explicitly models:
\[
P(S,A\mid X) =
\underbrace{\prod_{t=1}^{T}P_\theta(s_t \mid s_{<t}, E_s(x_{\rm ref}), E_s(x_{\rm mix}))}_{\text{Semantic Stage: tokens } S=(s_1,\dots,s_T)}
\times
\underbrace{\prod_{n=1}^{N}P_\phi(a_n \mid a_{<n}, S, E_a(x_{\rm ref}), E_a(x_{\rm mix}))}_{\text{Acoustic Stage: tokens } A=(a_1,\dots,a_N)}
\]
where $\theta$ and $\phi$ denote the transformer parameters at semantic and acoustic stages, respectively.

**Stage-1: Semantic Extraction**  
Inputs are continuous SSL embeddings $E_s(x) = \mathrm{WavLM}(x) \in \mathbb{R}^{T \times H}$ (from the 6th layer of WavLM). The model autoregressively produces $T$ discrete semantic tokens $\bar S = [\bar s_1, ..., \bar s_T]$.

**Stage-2: Acoustic Generation**  
Inputs comprise continuous codec embeddings $E_a(x) = \mathrm{DAC}(x) \in \mathbb{R}^{T \times H}$ (from a DAC encoder), together with semantic tokens $S$. The output sequence $\bar A = [\bar a_1, ..., \bar a_N]$ consists of fine-grained codec tokens produced by a single-codebook neural audio codec (“SimCodec”).

This hierarchical decomposition stabilizes generation and ensures semantic alignment between extracted speech and the reference [2512.20978].

## 2. Input Representations and Conditioning Mechanisms

GenTSE employs continuous, high-dimensional embeddings for conditioning at both stages, bypassing the need for projection layers:

- **Semantic embeddings** $E_s(x)$: Framewise outputs from WavLM (layer 6), providing robust context signals for semantic content.
- **Acoustic embeddings** $E_a(x)$: Framewise DAC encoder outputs, capturing low-level acoustic properties.
- The decoder-only transformers attend directly to these embeddings, integrating information from both the reference utterance and the input mixture.

This design delivers a richer context than previous discretized prompt methods or projection-based schemes and underpins the model’s ability to capture speaker- and content-specific characteristics [2512.20978].

## 3. Training Methodology: Reducing Exposure Bias and Optimizing Perceptual Quality

### 3.1 Baseline Training with Teacher Forcing
Cross-entropy losses are computed separately at each stage under teacher forcing:
\[
\mathcal{L}_{\rm CE,sem} = -\sum_{t=1}^T \log P_\theta(s_t|s_{<t}, E_s(x_{\rm ref}), E_s(x_{\rm mix}))
\]
\[
\mathcal{L}_{\rm CE,aco} = -\sum_{n=1}^N \log P_\phi(a_n|a_{<n}, S, E_a(x_{\rm ref}), E_a(x_{\rm mix}))
\]

### 3.2 Frozen-LM Conditioning (FLC) for Exposure Bias Mitigation
FLC addresses the mismatch between teacher-forcing during training and autoregressive inference (exposure bias):

1. Train baseline models $\theta$, $\phi$ under teacher forcing.
2. Clone parameters to $\theta' \leftarrow \theta$, $\phi' \leftarrow \phi$ and freeze the originals.
3. Generate token histories ($\bar S$, $\bar A$) using frozen models.
4. Train the clones ($\theta'$, $\phi'$) on their own predictions:
   \[
   \mathcal{L}'_{\rm CE,sem} = -\sum_{t=1}^T \log P_{\theta'}(\bar s'_t|\bar s_{<t}, E_s(x_{\rm ref}), E_s(x_{\rm mix}))
   \]
   \[
   \mathcal{L}'_{\rm CE,aco} = -\sum_{n=1}^N \log P_{\phi'}(\bar a'_n|\bar a_{<n}, S, E_a(x_{\rm ref}), E_a(x_{\rm mix}))
   \]

### 3.3 Direct Preference Optimization (DPO) for Human-Aligned Generation
To directly improve perceptual quality, GenTSE applies DPO to fine-tune the acoustic LM ($\psi$):

- For context $y=[S, E_a(x_{\rm ref}), E_a(x_{\rm mix})]$, top-$k$ samples ($M$ total) are scored by a pretrained UTMOS MOS predictor.
- For each $(A^+, A^-)$ pair ($A^+$ preferred), DPO loss is:
  \[
  \mathcal{L}_{\rm DPO} = -\mathbb{E}_{(A^+,A^-)} \left[ \log \sigma \left( \beta \log \frac{\pi_\psi(A^+ | y)\, \pi_{\rm ref}(A^- | y)}{\pi_{\rm ref}(A^+ | y)\, \pi_\psi(A^- | y)} \right) \right]
  \]
  where $\pi_{\psi}$ is the LM under optimization, $\pi_{\rm ref}$ is a frozen reference, and $\beta$ modulates sharpness.

This preference-integrated training directly steers generation towards human-valued speech characteristics [2512.20978].

## 4. Decoding and Inference Procedure

The hierarchical decoding pipeline comprises:

- **Semantic Decoding**: Greedy or beam search to select semantic token sequences $\bar S$ from $P_\theta$.
- **Acoustic Decoding**: Autoregressive top-$k$ sampling ($k=16$) of $\bar A$ from $P_\psi$; $M=32$ token sequences produced.
- **Candidate Selection**: For non-DPO models, best candidate is chosen by UTMOS score; for DPO-finetuned models, the single best sample is used.
- **Reconstruction**: Final selected sequence $\bar A$ is decoded into waveform $\hat x(t)$ via SimCodec.

This modular, sampling-based approach in both stages enables faithful, high-quality speech synthesis while preserving computational efficiency [2512.20978].

## 5. Experimental Results and Quantitative Comparison

GenTSE’s performance is evaluated on Libri2Mix (clean) against both discriminative (X-TF-GridNet, USEF-SepFormer) and generative LM-based (TSELM-L, LLaSE-G1, Metis) baselines. Evaluation encompasses speech quality, intelligibility, and speaker consistency using DNSMOS (SIG, BAK, OVRL), UTMOS, NISQA, dWER (Whisper-base), SpeechBERT, and SECS (Resemblyzer cosine similarity):

| Model                | SIG   | BAK   | OVRL  | UTMOS | NISQA | SECS  | dWER↓ | SpeechBERT |
|----------------------|-------|-------|-------|-------|-------|-------|-------|------------|
| GenTSE               | 3.656 | 4.135 | 3.399 | 4.296 | 3.976 | 0.928 | 0.177 | 0.920      |
| Metis (G)            | 3.588 | 3.980 | 3.265 | 3.882 | 3.869 | 0.879 | 0.180 | 0.890      |
| LLaSE-G1 (G)         | 3.531 | 4.015 | 3.226 | 3.228 | 3.638 | 0.839 | 0.476 | 0.825      |
| TSELM-L (G)          | 3.478 | 4.035 | 3.198 | 3.556 | 3.509 | 0.651 | 0.263 | 0.832      |
| USEF-SepFormer (D)   | 3.324 | 3.698 | 2.927 | 3.492 | 2.880 | 0.806 | 0.156 | 0.830      |
| Mixture              | 3.383 | 3.098 | 2.653 | 1.519 | 2.251 | 0.754 | 0.821 | 0.655      |

GenTSE achieves superior overall performance, with improvements over Metis in OVRL (+0.134), SECS (+0.049), and dWER ($\approx$0.177), signifying advances in both linguistic and speaker-specific fidelity [2512.20978].

## 6. Qualitative Insights, Ablations, and Methodological Contributions

**Hierarchical Design**: Ablation studies removing the semantic stage result in higher dWER (0.284→0.217), empirically validating the coarse-to-fine arrangement.

**Frozen-LM Conditioning**: Training with FLC closes the accuracy gap between teacher forcing and autoregressive inference, indicating effective mitigation of exposure bias (see Fig. 2 in [2512.20978]).

**Perceptual Alignment via DPO**: DPO fine-tuning increases UTMOS by $\approx0.259$ (4.384 vs. 4.125) after 400 steps, aligning outputs with human judgments. This improvement comes with minimal trade-off in log-likelihood-based metrics.

**Integrated Generative Pipeline**: The unified LM-based strategy—leveraging continuous conditioning, FLC, and DPO—surpasses previous LM-based TSE models without introducing extra inference-time modules, yielding more natural, intelligible, and speaker-consistent extracted speech [2512.20978].

A plausible implication is that the hierarchical, preference-aligned generative modeling paradigm found in GenTSE may generalize to related sequence-generation tasks within speech and multimodal domains.

Source: https://www.emergentmind.com/topics/gentse