Replace gr.Audio outputs with gr.Textbox to return CDN URLs directly

Add verbose R2 upload logging for debugging

Add R2 upload: serve audio via cdn.koevoice.com/temp_audios/ (no more 403/500)

Add R2 upload: serve audio via cdn.koevoice.com/temp_audios/ (no more 403/500)

Fix URL audio download: add User-Agent header

Fix UI: move generate_unified outside gr.Blocks context

Remove allowed_paths parameter from demo.launch

Complete UI overhaul: single unified interface

Add URL audio download with caching

Fix remaining dataframe references after UI change

Replace Dataframe with simple Textbox for text input

Show 3 rows by default to make +/- buttons visible

Add UI hint for adding rows to text dataframe

Revert Simple tab to use original sample_pipeline

Convert speaker_latent to model dtype

Fix UnboundLocalError in model forward

Revert to original approach: reuse speaker encoding, not KV cache

Build speaker cache fresh in each @spaces.GPU call

Ensure tensor contiguity before concatenation

Fix k_norm application in KV cache building

Fix mask handling for prebuilt KV cache

Fix PCA state loading for ZeroGPU compatibility

Fix PCA state repo path to jordand/echo-tts-base

Fix ZeroGPU compatibility: always load to CPU first

Add batch text generation with speaker cache optimization

Update app.py

Update app.py

add simple version

Update text_presets.txt

Update app.py

Update inference.py

use newer model

Update requirements.txt

Update inference.py

Update app.py

Update README.md

Update README.md

Update app.py

Update silentcipher/server.py

Update requirements.txt

Update requirements.txt

Update app.py

Update requirements.txt

Upload 21 files

Update README.md

initial commit