[Explainer] Probing Low Frame Rate Degration in Neural Audio Codecs

Created on June 26, 2026

2026   ·   tts   genai   ·   technical

This is the first post in the technical stream of the blog — speech, NLP, and ML, with a bias toward African and other under-resourced languages.

The theme supports everything I need for these write-ups out of the box:

  • LaTeX math via MathJax, e.g. the CTC objective \(\mathcal{L}_{\text{CTC}} = -\log \sum_{\pi \in \mathcal{B}^{-1}(y)} p(\pi \mid x)\).

  • Syntax-highlighted code:

    import torch
from espnet2.bin.asr_inference import Speech2Text

s2t = Speech2Text.from_pretrained("espnet/your_model")
text, *_ = s2t(speech)[0]
print(text)
  • Jupyter notebook embeds (see the al-folio docs for the jupyter include).

More to come on ASR, TTS, machine translation, and speaker representations.