| --- |
| title: Textual Echo Cancellation (TEC) |
| emoji: 🎙️ |
| colorFrom: blue |
| colorTo: indigo |
| sdk: static |
| pinned: false |
| license: apache-2.0 |
| models: |
| - wq2012/tec_single_interfering |
| - wq2012/tec_multi_interfering |
| - wq2012/aec_single_interfering |
| - wq2012/aec_multi_interfering |
| - wq2012/vanilla_seq2seq_single_interfering |
| - wq2012/vanilla_seq2seq_multi_interfering |
| short_description: Cancel TTS playback echo using source text and compare ASR |
| --- |
| |
| # Textual Echo Cancellation (TEC) — Interactive Demo |
|
|
| This Space demonstrates **Textual Echo Cancellation (TEC)** based on the IEEE SLT 2021 paper: |
|
|
| > **Textual Echo Cancellation** |
| > *Shaojin Ding, Ye Jia, Ke Hu, Quan Wang* |
| > - **Paper**: [arXiv:2008.06006](https://arxiv.org/abs/2008.06006) |
| > - **GitHub Repository**: [https://github.com/wq2012/tec](https://github.com/wq2012/tec) |
| > - **PyPI Package**: [`textual-echo-cancellation`](https://pypi.org/project/textual-echo-cancellation/) (`pip install textual-echo-cancellation`) |
| > - **Pretrained Models**: [`wq2012/tec_single_interfering`](https://huggingface.co/wq2012/tec_single_interfering) | [`wq2012/tec_multi_interfering`](https://huggingface.co/wq2012/tec_multi_interfering) | [`wq2012/aec_single_interfering`](https://huggingface.co/wq2012/aec_single_interfering) |
| > - **Audio Demo Page**: [https://google.github.io/speaker-id/publications/TEC/](https://google.github.io/speaker-id/publications/TEC/) |
|
|
| ## How It Works |
|
|
| 1. **Upload an audio file** containing a microphone mixture (a user speaking while a device is playing back a Text-to-Speech response) or select one of the 4 built-in benchmark examples. |
| 2. **Enter the interfering TTS text** (the source text of the device's TTS playback) and select a model (`TecSingleInterfering`, `TecMultiInterfering`, or `AecSingleInterfering`). |
| 3. Click **Run Textual Echo Cancellation & Compare ASR**: |
| - Runs **ASR on the user uploaded audio file** (Before TEC). |
| - Runs **Echo Cancellation** (loading the exact Lingvo checkpoint outputs in `examples/sample_1..4_{tec_single,tec_multi,aec_single}.wav` generated by [`generate_examples.py`](generate_examples.py) for the built-in benchmark examples, or running an in-browser WebAudio text-conditioned spectral canceller for custom uploaded files). |
| - Runs **ASR on the Echo Cancelled audio file** (After TEC) and prints out the recognition results for both approaches side by side. |
|
|
| *(This repository also includes [`generate_examples.py`](generate_examples.py) and the standalone Python Gradio app in [`app.py`](app.py) + [`requirements.txt`](requirements.txt) for full Lingvo/TensorFlow checkpoint inference on arbitrary files via `python3 app.py`.)* |
|
|
