Speak-n-Spell Analysis & Custom Voice Pack Software

Enables analysis and modification of the speech sounds on 2019 Speak&Spell toys (sold under the "Basic Fun" brand with "Classic 80's Design").

Installation / Prerequisites

Python environment

To install the spana (speak-and-spell analysis) package into a virtual environment, run:

cd speaknspell_analysis/
python3.13 -m venv .venv
source .venv/bin/activate

pip install -U pip

# install `spana` from this checkout in editable mode
pip install -e .

Other (optional) Tools

• em100 open-source command-line tool
  • from here
  • Compile and install it so that the em100 binary is available on your PATH
  • Remember to install udev rules to /etc/udev/rules.d/
• flashrom (for reading and re-programming your SPI flash)
  • from here
  • Compile and install it so that the flashrom command is available on your PATH

Usage

There are 3 main capabilities provided in this repository:

1. Compiling a new Speak&Spell 2019-compatible "voice pack."
2. Decoding a SPI Flash Image
3. "Live Tracing" of SPI Flash data access with python -m spana.live_trace

Note: 2 and 3 require that you have an extracted Speak&Spell 2019 Flash Image (see below for how to do this with a SPI Flash reader).

Compiling a new voice pack.

You will need 223 or so numerically-named WAV files (at 10kHz sample rate, currently the code does not do resampling). You can generate these yourself (by recording yourself saying the words and phrases in known_phrases.csv) or use the existing ones in the voice_packs/snoop_dogg directory.

To compile:

$ python -m spana.compile_voice_pack --wav-dir voice_packs/snoop_dogg/
Encoding 224 files in 14 parallel processes...
Allocating...
Relocating...
Saving to compiled_voice_pack.bin

Note: The script looks for a 3-digit number (with leading zeros) in the wav filename to map the file to a particular offset table index. The 3-digit number just has to be somewhere in the filename, before the .wav suffix (e.g. angel_042.wav, 042_angel.wav, and ANG042EL.wav are all acceptable).

Usage:

python -m spana.compile_voice_pack --help
usage: compile_voice_pack [-h] -d WAV_DIR

Build a Speak & Spell flash image from wav files

options:
  -h, --help            show this help message and exit
  -d, --wav-dir WAV_DIR
                        Input search dir for wav files

The directory specified with -d/--wav-dir should contain 223 WAV files (at 10kHz sample rate). Each wav file should have a 3-digit (with leading zeros) number in the filename
(e.g. 042_angel.wav) which indicates which record in the offset table it will be encoded into. See known_phrases.csv for a list of all known Speak&Spell words.

decoder

Decodes all of the sound blobs from a 2019 Speak&Spell SPI Flash image into wav files.

Operates on the ORIGINAL_FLASH_IMAGE file by default, or can be specified on command line (run with --help for options)

Example:

$ python -m spana.decode_sounds_to_wav
  wrote decoded_sounds/ss_000_A.wav, enc size:  2944, pcm size: 11178, CR: 3.80
  wrote decoded_sounds/ss_001_B.wav, enc size:  2448, pcm size: 9292, CR: 3.80
  wrote decoded_sounds/ss_002_C.wav, enc size:  2784, pcm size: 10580, CR: 3.80
  wrote decoded_sounds/ss_003_D.wav, enc size:  2192, pcm size: 8188, CR: 3.74
  wrote decoded_sounds/ss_004_E.wav, enc size:  2368, pcm size: 8924, CR: 3.77
  wrote decoded_sounds/ss_005_F.wav, enc size:  2128, pcm size: 8004, CR: 3.76
...

live_trace (requires em-100 SPI flash emulator)

Loads an (optionally modified) SPI flash image onto the EM-100 SPI Flash Emulator

Example:

$ python -m spana.live_trace -M chipmunk_mode

Extracting a flash image from your own Speak & Spell SPI Flash (for use with live_trace)

Due to intellectual property concerns, I do not distribute the 2019 Speak&Spell SPI Flash image here.

However, you can extract your own image from your Speak&Spell using a SPI flash reader. A good cheap option is the CH341A SPI Flash reader, which is readily available on Amazon, eBay, aliexpress, etc.

The CH341A is compatible with the venerable flashrom command on Linux, and modern versions of flashrom command support the BoyaMicro SPI flash IC I found inside my unit (although it's possible that Speak&Spells have been produced with other SPI flash chips as well).

cd speaknspell_analysis
mkdir -p flash_images
sudo flashrom -p ch341a_spi -c 'XM25QH80B' -r flash_images/ORIGINAL_FLASH_IMAGE

Code in this repository which depends on the original flash image assumes you'll place it in speaknspell_analysis/flash_images/ORIGINAL_FLASH_IMAGE.

The sha256 sum of the flash image I analyzed previously is: 0da032eae0bd5665ad4f3905d9e16f5b7efc75f04907b26e79f96e74380ce511. If your flash image has a different hash, perhaps you've discovered a new version (or check your flash reader for hookup issues).

I just want Snoop and Spell

To convert your 2019 Speak & Spell into a Snoop & Spell:

1. Desolder the SPI Flash IC
2. Insert into CH341A flash reader
3. (optional) Make a backup as described above in "Extracting a flash image" (you never know)
4. Run the following command (with a recent version of flashrom) to write the Snoop & Spell voice pack to the flash IC:

flashrom -p ch341a_spi -c 'XM25QH80B' -w snoop_and_spell.bin

5. Re-solder the flash IC into place.
6. Be the envy of all your friends.

Related Presentations

• As presented at DistrictCon Year 0 (22 Feb 2025): Hacking Childhood Memories A tale of reverse engineering, audio encodings, and never growing up
  • Presentation Slides
  • Presentation Video