I desoldered the flash chip from my Russian-songs version of the Dancing Cactus and dumped it using an ESP32. Like the example on Leo Leung’s wiki, the firmware contains the SH54 signature and a similar SPI bootloader layout.

However, in the site's dump, the start of the first audio segment at offset 0x0BB000 has a WAV header (5741 5604), while in mine it starts with the MP3 sync header: FF FB — confirming this unit uses MP3 audio instead of WAV.

Here’s the full dump for comparison or analysis:
https://filebin.net/eo8low45btuq9m6c

Let me know if it's helpful to extract and convert the MP3s from it. Happy to contribute more!