RYM AudioBook English and Spanish Version Exciting UPDATE

After attempting to upload the Spanish RYM audiobook to audio platforms such as Google, Apple, Spotify, and Audible, we had to revisit the project that Yvonne and I first tried to submit to Audible many months ago, while the audiobook was being published in English. Audible recently responded with the following:

"Problem: The title contains audio produced using an unauthorized text-to-speech (TTS) system. This does not comply with ACX requirements.

According to ACX requirements (https://help.acx.com/s/article/what-are-the-acx-audio-submission-requirements#a), the audiobook you submit must be narrated by a human, unless otherwise authorized."

If you personally recorded the chapters, then the issue is no longer “Is this AI?”, but rather “Why is ACX classifying it as AI?”

This is happening more and more often on audiobook platforms. Automated detection systems can generate false positives for several reasons:

• Excessive voice processing (aggressive noise reduction, de-essing, compression, EQ).
• Removing nearly all breaths from the recording.
• Very uniform editing between sentences and pauses.
• Recording with extremely clean microphones or professional audio chains.
• Using restoration tools such as iZotope RX, Adobe Enhance Speech, Descript, Auphonic, etc.
• Extreme consistency in tone and volume throughout hours of narration.

ACX generally does not reveal exactly what triggered the alert, and their detection systems are not infallible.

ACX does not state that they have proven AI was used. Rather, they state that their Quality Assurance team determined that the title "contains audio produced using an unauthorized TTS system." This could be the result of an automated detection process or a human review based on characteristics they believe are consistent with text-to-speech narration.

One detail that stands out is that the audio was processed using Auphonic.

Although Auphonic is not a text-to-speech or voice-cloning system, it does apply sophisticated audio processing, including loudness normalization, noise reduction, dynamic range control, and other mastering enhancements. If the original recording was already very clean and, in addition:

• Breaths were removed or significantly reduced,
• Pauses were edited to be highly consistent,
• All chapters were processed using the same settings,

the final result can sound surprisingly similar to a modern synthetic voice.

Based on the information available, my assessment is that this is likely a false positive.

Rather than re-recording all 17 chapters, I would recommend appealing the decision immediately.

The strongest points in support of the appeal are:

• The narration is entirely Anastasia's own voice.
• No text-to-speech technology or voice cloning was used.
• Auphonic was used only for mastering and audio enhancement.
• Original, unprocessed recordings are still available for at least some chapters.
• We are willing to provide raw audio samples for verification.

In fact, having even two or three original chapters may be sufficient to demonstrate that the narration was performed by a human narrator and that the final audiobook is derived from those authentic recordings.

----------------------

In this case, I believe there may have been too much audio processing involved. The files I later processed with Auphonic to remove residual background noise had already been processed in GarageBand for the YouTube version.