Help: High-speed playback sounds garbled on complex jargon.

I've been conducting a systematic evaluation of several text-to-speech (TTS) solutions for my team's research consumption workflow, with a particular focus on high-speed playback for skimming academic papers and technical documentation. Speechify's premise is compelling, but I'm encountering a significant, reproducible issue that undermines its utility for my primary use case: at playback speeds above ~3x, complex jargon—specifically multi-syllabic technical terms, proper nouns in computer science, and dense medical or legal terminology—becomes garbled and often unintelligible.

The degradation isn't linear. At 2.5x, clarity is generally acceptable. Crossing the 3x threshold introduces artifacts that seem to be a failure of the speech synthesis model to properly handle coarticulation in unfamiliar word combinations. It sounds less like sped-up speech and more like a phoneme-level compression error.

**My testing methodology was as follows:**
* **Sample Texts:** I used three standardized passages:
1. A section from a machine learning paper featuring terms like "stochastic gradient descent," "non-parametric Bayesian inference," and "convolutional neural network."
2. A clinical trial abstract with drug names (e.g., "adalimumab," "tocilizumab") and medical conditions.
3. A legal contract excerpt with Latin terms and compounded nouns.
* **Control:** The same passages were tested at 1x (baseline), 2x, 2.5x, 3x, and 3.5x speeds.
* **Voice Variable:** Tested across multiple "premium" voices (Snoop, Gwyneth, and a default English voice).
* **Environment:** Desktop application (macOS), latest version, with no other CPU-intensive processes running.

The results were consistent across voices. The system appears to be performing simple time-domain compression on a generated audio stream, rather than having a speed-optimized synthesis pipeline that adjusts phoneme duration and intonation models for high-velocity output. This is a known challenge in TTS, but competing tools (like NaturalReader's professional engine) handle it better, albeit with other trade-offs.

I'm seeking insights from other users who leverage Speechify for technical material:
* Is this a recognized limitation of the current engine?
* Are there specific voice or playback settings (e.g., enabling/disabling certain processing features) that mitigate this, or does the web app perform differently than the desktop?
* Has anyone performed a comparative A/B test on jargon intelligibility at speed against other high-end TTS services?

My hypothesis is that the underlying model is optimized for conversational speed and vocabulary, and the high-speed playback is a post-processing feature, not a core synthesis capability. I'd like to validate this before finalizing my tooling recommendation.