CAT-1000 LPC Speech Extraction

Extract, synthesize, and label all spoken phrases from the CAT-1000 repeater controller's voice EPROM.

Device: CAT-1000 Repeater Controller, firmware V3.04A (1998) Speech chip: TI TSP53C30 (TMS5220-compatible LPC-10 vocoder) Result: 481 labelled phrases as .lpc (raw bitstream) and .wav (synthesized audio) — vocabulary complete

---

Quick Start

# Install PyTI_LPC_CMD into a local subdirectory (required for --wav only)
bash INSTALL_PyTI.sh
# Clones https://github.com/ke4ahr/PyTI_LPC_CMD into PyTI_LPC_CMD/

# Export LPC files to cat-1000_lpc_clips/ (no PyTI dependency)
python3 cat-1000_lpc_export.py cat-1000_clips.csv -o cat-1000_lpc_clips/

# Export LPC + WAV to separate directories (requires PyTI_LPC_CMD)
python3 cat-1000_lpc_export.py cat-1000_clips.csv --wav \
    -o cat-1000_lpc_clips/ --wav-dir cat-1000_wav_clips/

Requires: Python 3.8+. PyTI_LPC_CMD (for --wav) cloned into PyTI_LPC_CMD/ by INSTALL_PyTI.sh.

---

Files

File	Description
eprom_images/cat-1000-voice_27SF512.BIN	Voice EPROM image (64 KB) — primary input
eprom_images/cat-1000-V304A_program_27C512.BIN	Program EPROM image (64 KB) — reference
cat-1000_clips.csv	Primary phrase table (no header): NNN,"Label",0xSTART,0xEND -- 481 entries, 3-digit zero-padded word number. Input to cat-1000_lpc_export.py; read/rewritten by fix_csv_addresses.py.
cat-1000_phrases.csv	Alternate phrase table (has header row): word_number,word_text,start_address,end_address,notes -- same 481 phrases, sorted alphabetically by label. Used by cat-1000_lpc_export.py --dump-csv to populate word labels when regenerating cat-1000_clips.csv.
cat-1000_lpc_export.py	Main export script
cat-1000_analysis.py	EPROM analysis utilities
cat-1000_lpc_repack.py	Repack modified LPC data into EPROM format
INSTALL_PyTI.sh	Clone PyTI_LPC_CMD into PyTI_LPC_CMD/ subdirectory
PyTI_LPC_CMD/	PyTI_LPC_CMD synthesizer (cloned by INSTALL_PyTI.sh; not in repo)
cat-1000_lpc_clips/	Output: NNNN_Label.lpc for each phrase
cat-1000_wav_clips/	Output: NNNN_Label.wav for each phrase (synthesized)
build_docs.py	Build all documentation PDFs from source
docs/	Technical documentation

---

Voice EPROM Structure

0x0000–0x009E  8086 dispatch routine
0x009F–0x0641  Phrase index (9 groups × N entries, 3 bytes each)
0x0642–0xC4C1  TMS5220 LPC bit-packed speech data (~49 KB)

The phrase index is self-contained in the voice EPROM. The program EPROM calls CALL FAR F000:0000 with AH=group_id, AL=phrase_id; the voice EPROM dispatch code performs a linear search and returns BX=start_offset.

Word numbering: word_number = group * 100 + phrase_id

---

Phrase Boundary Method

End addresses are not stored in the EPROM index. Each phrase ends where the next phrase (sorted by start address) begins. This is authoritative and used exclusively in cat-1000_clips.csv.

sorted_starts = sorted(set(addr for each index entry))
end_of_phrase = sorted_starts[i+1]   # or 0xC4C1 for the last phrase

---

Bit Direction

LPC bytes are stored native MSB-first: bit 7 of each byte is the MSB of the first LPC parameter field. Read bytes in EPROM order, process each byte from bit 7 down to bit 0. Do not bit-reverse before parsing.

The export script handles a double-reversal internally to match PyTI's synthesis convention — this is intentional and correct.

---

CSV Format

cat-1000_clips.csv (no header row):

NNN,"Label",0xSTART,0xEND

cat-1000_phrases.csv (has header):

word_number,word_text,start_address,end_address,notes

All addresses are verified against the voice EPROM phrase index as of 2026-04-02.

---

Export Script Options

python3 cat-1000_lpc_export.py [CSV] [options]

  --wav           Synthesize WAV files (requires PyTI_LPC_CMD)
  -o DIR          Output directory for .lpc files (default: lpc_phrases/)
  --wav-dir DIR   Separate output directory for .wav files (default: same as -o)
  --voice FILE    Voice EPROM image (default: eprom_images/cat-1000-voice_27SF512.BIN)
  -q              Suppress per-file output
  --dump-csv      Print phrase table as CSV to stdout (no input CSV needed)

---

CAT-310DX — Digital Weather Station a.k.a. Road Monitor

Device: CAT-310DX V1.00 (C)1998, checksum 7A69 CPU: Intel 8052 (MCS-51). Speech chip: TSP53C30 (TMS5220-compatible) at MOVX 0xC000. Result: 414 unique LPC clips extracted and synthesized — numbers 0–99, 4–9 recordings each

# Extract all 414 clips (LPC + WAV)
python3 cat-310dx_extract.py --wav
# LPC: cat-310dx_lpc_clips/  WAV: cat-310dx_wav_clips/  CSV: cat-310dx_clips.csv

File	Description
eprom_images/CAT-310DX_V1-00_1998_7A69.BIN	Unified firmware + speech EPROM (64 KB)
cat-310dx_extract.py	Extraction script -- reads dispatch table at 0x4AC3, writes sequence-numbered clips
cat-310dx_correlate.py	Correlate 310DX clips against CAT-1000 clips by checksum
cat-310dx_rename_clips.py	Rename old-style address-named clips using correlation map
cat-310dx_synth.py	Bit-direction validation tool
cat-310dx_lpc_clips/	Output: 414 × .lpc clips
cat-310dx_wav_clips/	Output: 414 × .wav clips (synthesized)
cat-310dx_clips.csv	Phrase table: seq number, label, start/end address (414 entries)
docs/cat310dx_re_report.md	Full reverse engineering report
docs/cat310dx_analysis.md	Technical reference: memory map, ISR disassembly, dispatch table

LPC format is identical to CAT-1000 (MSB-first TMS5220 codec). The same synthesis pipeline handles both devices.

# Re-extract all 414 clips with sequence-numbered names
python3 cat-310dx_extract.py --wav
# LPC: cat-310dx_lpc_clips/  WAV: cat-310dx_wav_clips/

# Correlate against CAT-1000 to find byte-identical clips
python3 cat-310dx_correlate.py

---

Documentation

• docs/cat-1000_re_report.md — CAT-1000 reverse engineering report
• docs/cat310dx_re_report.md — CAT-310DX reverse engineering report
• docs/lpc_extraction_process.md — Complete LPC extraction process: EPROM layout, phrase index, bitstream format, synthesis pipeline
• docs/handoff.md — Project state and known pitfalls for both devices
• docs/csv_address_audit.md — History of the CAT-1000 phrase boundary audit and corrections
• docs/activity_log.md — Work log 2026-03-23 to 2026-04-05

Building Documentation

python3 build_docs.py

Compiles all LaTeX papers, TikZ diagrams, Graphviz diagrams, and Markdown documents to PDF. Requires pdflatex, xelatex, dot (Graphviz), and pandoc. See build_docs.md for details.

---

Verification

After export, cat-1000_wav_clips/0220_Affirmative.wav should be ~875 ms and sound clearly like "Affirmative." cat-1000_lpc_clips/0220_Affirmative.lpc should be 180 bytes.

---

Acknowledgements

The author wishes to thank the following: Al Preuss W4MGK (SK), Ron Eichholtz WB4UFA, Harlie Henson KB4CRG, Brian Kirby KD4FM, Russ Harper, Marion "Gib" Gibson W4LHR (SK), Leigh Bartlow WD4CPF, The Huntsville Amateur Radio Club (K4BFT), Ron Shaffer W4VM (SK), Ralph Hogan W4XE (ex: WB4TUR), The North Alabama Repeater Association (NARA), Don Hediger N4MSN, The Marshall Space Flight Center Amateur Radio Club NN4SA (ex: WA4NZD), Robert Meister WA1MIK (SK), Mike Morris WA6ILQ, Josh Hatton W4ZZK, Kevin K. Custer W3KKC, CAT-Controllers at groups.io, catauto at groups.io, Sig Loeb W4WOH (SK), and the various named, unnamed, pseudonymous, and anonymous members of the TI-99/4A, TI Speech, CAT Auto, and Repeater-Builder communities.

---

Copyright (C) 2026 Kris Kirby, KE4AHR. This document is licensed under the GNU General Public License v3.0 or later. You may redistribute and/or modify it under the terms of the GNU GPL as published by the Free Software Foundation. See https://www.gnu.org/licenses/ for details.