diff --git a/firmware/boards/buzzy_nrf52840.conf b/firmware/boards/buzzy_nrf52840.conf
index 6e54c67..6126862 100644
--- a/firmware/boards/buzzy_nrf52840.conf
+++ b/firmware/boards/buzzy_nrf52840.conf
@@ -6,3 +6,6 @@ CONFIG_RTT_CONSOLE=y
 CONFIG_UART_CONSOLE=n
 CONFIG_LOG_BACKEND_UART=n
 CONFIG_LOG_BACKEND_RTT=y
+
+# Keep SPI NOR page layout aligned with generated LittleFS block size (4KB).
+CONFIG_SPI_NOR_FLASH_LAYOUT_PAGE_SIZE=4096
diff --git a/firmware/fs/.gitignore b/firmware/fs/.gitignore
new file mode 100644
index 0000000..fb8101c
--- /dev/null
+++ b/firmware/fs/.gitignore
@@ -0,0 +1,3 @@
+lfs_external_flash.hex
+.tmp_lfs_build/
+*.wav
diff --git a/firmware/fs/build_lfs_audio.py b/firmware/fs/build_lfs_audio.py
new file mode 100644
index 0000000..e2d55a5
--- /dev/null
+++ b/firmware/fs/build_lfs_audio.py
@@ -0,0 +1,221 @@
+#!/usr/bin/env python3
+import argparse
+import shutil
+import subprocess
+import tempfile
+from pathlib import Path
+
+import soundfile as sf
+import yaml
+from intelhex import IntelHex
+from kokoro import KPipeline
+from littlefs import LittleFS
+
+DEFAULT_SAMPLE_RATE = 16000
+DEFAULT_VOICE = "af_bella"
+DEFAULT_BLOCK_SIZE = 4096
+DEFAULT_BLOCK_COUNT = 2048
+DEFAULT_FILTERS = [
+    "highpass=f=120",
+    "lowpass=f=6000",
+    "acompressor=threshold=-18dB:ratio=3:attack=5:release=80",
+    "loudnorm=I=-16:TP=-1.0",
+]
+
+DEFAULT_SYS_PROMPTS = [
+    {"id": "404", "text": "No sound sample was found on the device."},
+    {"id": "update", "text": "Firmware updated. Awaiting confirmation."},
+    {"id": "confirm", "text": "State confirmed."},
+]
+
+
+def run_ffmpeg(cmd: list[str]) -> None:
+    subprocess.run(cmd, check=True, stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL)
+
+
+def export_raw_pcm(input_path: Path, output_path: Path, sample_rate: int, filters: list[str]) -> None:
+    output_path.parent.mkdir(parents=True, exist_ok=True)
+    filter_str = ",".join(filters)
+    run_ffmpeg([
+        "ffmpeg",
+        "-y",
+        "-i",
+        str(input_path),
+        "-af",
+        filter_str,
+        "-ar",
+        str(sample_rate),
+        "-ac",
+        "1",
+        "-f",
+        "s16le",
+        "-acodec",
+        "pcm_s16le",
+        str(output_path),
+    ])
+
+
+def load_sys_prompts(yaml_path: Path | None) -> list[dict]:
+    if yaml_path is None:
+        return DEFAULT_SYS_PROMPTS
+
+    with open(yaml_path, "r", encoding="utf-8") as f:
+        data = yaml.safe_load(f) or []
+
+    prompts: list[dict] = []
+    if isinstance(data, list):
+        for item in data:
+            if isinstance(item, dict) and item.get("id") and item.get("text"):
+                prompts.append({"id": item["id"], "text": item["text"]})
+    elif isinstance(data, dict):
+        for item in data.get("assets", []):
+            if isinstance(item, dict) and item.get("id") and item.get("text"):
+                prompts.append({"id": item["id"], "text": item["text"]})
+
+    return prompts
+
+
+def build_sys_tts(prompts: list[dict], out_sys_dir: Path, sample_rate: int) -> int:
+    pipeline = KPipeline(lang_code="a")
+    generated = 0
+
+    with tempfile.TemporaryDirectory(prefix="tts_tmp_") as tmp_dir_name:
+        tmp_dir = Path(tmp_dir_name)
+
+        for prompt in prompts:
+            asset_id = prompt["id"]
+            text = prompt["text"]
+
+            tmp_wav = tmp_dir / f"{asset_id}.wav"
+            generator = pipeline(text, voice=DEFAULT_VOICE, speed=1.0)
+            for _, _, audio in generator:
+                sf.write(tmp_wav, audio, 24000)
+                break
+
+            out_file = out_sys_dir / asset_id
+            export_raw_pcm(tmp_wav, out_file, sample_rate=sample_rate, filters=DEFAULT_FILTERS)
+            generated += 1
+
+    return generated
+
+
+def import_wavs_to_a(wav_dir: Path | None, out_a_dir: Path, sample_rate: int) -> int:
+    if wav_dir is None:
+        return 0
+    if not wav_dir.exists() or not wav_dir.is_dir():
+        raise FileNotFoundError(f"WAV directory not found: {wav_dir}")
+
+    count = 0
+    for wav_file in sorted(wav_dir.glob("*.wav")):
+        out_file = out_a_dir / wav_file.stem
+        export_raw_pcm(wav_file, out_file, sample_rate=sample_rate, filters=DEFAULT_FILTERS)
+        count += 1
+
+    return count
+
+
+def add_files_recursive(fs: LittleFS, local_path: Path, lfs_path: str = "/") -> None:
+    if not local_path.exists():
+        return
+
+    for item in sorted(local_path.iterdir()):
+        target_path = f"{lfs_path.rstrip('/')}/{item.name}".replace("//", "/")
+        if item.is_file():
+            with open(item, "rb") as f:
+                with fs.open(target_path, "wb") as lfs_file:
+                    lfs_file.write(f.read())
+        elif item.is_dir():
+            fs.mkdir(target_path)
+            add_files_recursive(fs, item, target_path)
+
+
+def build_littlefs_hex(source_folder: Path, output_hex: Path, block_size: int, block_count: int, start_addr: int) -> None:
+    fs = LittleFS(
+        block_size=block_size,
+        block_count=block_count,
+        read_size=256,
+        prog_size=256,
+        lookahead_size=512,
+        cache_size=4096,
+    )
+
+    add_files_recursive(fs, source_folder)
+
+    ih = IntelHex()
+    lfs_buffer = fs.context.buffer
+
+    for idx in range(block_count):
+        offset = idx * block_size
+        block_data = lfs_buffer[offset : offset + block_size]
+        if any(byte != 0xFF for byte in block_data):
+            ih.frombytes(block_data, offset=start_addr + offset)
+
+    ih.tofile(str(output_hex), format="hex")
+
+
+def parse_args() -> argparse.Namespace:
+    parser = argparse.ArgumentParser(
+        description=(
+            "Generate Bella-only TTS in /lfs/sys, optionally import WAVs into /lfs/a, "
+            "then create LittleFS HEX."
+        )
+    )
+    parser.add_argument("--wav-dir", default=None, help="Optional directory with WAV files for /lfs/a")
+    parser.add_argument("--sys-yaml", default=None, help="Optional YAML with system prompts (id/text)")
+    parser.add_argument("--sample-rate", type=int, default=DEFAULT_SAMPLE_RATE)
+    parser.add_argument("--block-size", type=int, default=DEFAULT_BLOCK_SIZE)
+    parser.add_argument("--block-count", type=int, default=DEFAULT_BLOCK_COUNT)
+    parser.add_argument("--start-addr", type=lambda x: int(x, 0), default=0x12000000)
+    parser.add_argument(
+        "--output-hex",
+        default=str(Path(__file__).resolve().parent / "lfs_external_flash.hex"),
+        help="Output path for generated HEX",
+    )
+    parser.add_argument(
+        "--keep-staging",
+        action="store_true",
+        help="Keep temporary staging directory for inspection",
+    )
+    return parser.parse_args()
+
+
+def main() -> None:
+    args = parse_args()
+
+    fs_dir = Path(__file__).resolve().parent
+    output_hex = Path(args.output_hex).resolve()
+    wav_dir = Path(args.wav_dir).resolve() if args.wav_dir else None
+    sys_yaml = Path(args.sys_yaml).resolve() if args.sys_yaml else None
+
+    staging_root = fs_dir / ".tmp_lfs_build"
+    if staging_root.exists():
+        shutil.rmtree(staging_root)
+
+    out_sys = staging_root / "lfs" / "sys"
+    out_a = staging_root / "lfs" / "a"
+    out_sys.mkdir(parents=True, exist_ok=True)
+    out_a.mkdir(parents=True, exist_ok=True)
+
+    prompts = load_sys_prompts(sys_yaml)
+    if not prompts:
+        raise RuntimeError("No system prompts available. Provide --sys-yaml or use defaults.")
+
+    tts_count = build_sys_tts(prompts, out_sys, args.sample_rate)
+    wav_count = import_wavs_to_a(wav_dir, out_a, args.sample_rate)
+
+    build_littlefs_hex(
+        source_folder=staging_root,
+        output_hex=output_hex,
+        block_size=args.block_size,
+        block_count=args.block_count,
+        start_addr=args.start_addr,
+    )
+
+    print(f"Done. TTS assets: {tts_count}, WAV imports: {wav_count}, HEX: {output_hex}")
+
+    if not args.keep_staging and staging_root.exists():
+        shutil.rmtree(staging_root)
+
+
+if __name__ == "__main__":
+    main()
diff --git a/firmware/fs/program.bat b/firmware/fs/program.bat
index 2a210a8..99640ce 100644
--- a/firmware/fs/program.bat
+++ b/firmware/fs/program.bat
@@ -1 +1 @@
-nrfutil device --x-ext-mem-config-file buzzy.json program --firmware \\10.0.10.30\edi\nrf_playground\lasertag\firmware\tools\littlefs_generator\lfs_external_flash.hex --options verify=VERIFY_READ,reset=RESET_SYSTEM
\ No newline at end of file
+nrfutil device --x-ext-mem-config-file "%~dp0buzzy.json" program --firmware "%~dp0lfs_external_flash.hex" --options verify=VERIFY_READ,reset=RESET_SYSTEM
\ No newline at end of file
diff --git a/firmware/fs/program.sh b/firmware/fs/program.sh
new file mode 100755
index 0000000..0af3fb3
--- /dev/null
+++ b/firmware/fs/program.sh
@@ -0,0 +1,4 @@
+#!/usr/bin/env sh
+SCRIPT_DIR="$(CDPATH= cd -- "$(dirname -- "$0")" && pwd)"
+
+nrfutil device --x-ext-mem-config-file "$SCRIPT_DIR/buzzy.json" program --firmware "$SCRIPT_DIR/lfs_external_flash.hex" --options verify=VERIFY_READ,reset=RESET_SYSTEM
diff --git a/firmware/fs/requirements.txt b/firmware/fs/requirements.txt
new file mode 100644
index 0000000..f543831
--- /dev/null
+++ b/firmware/fs/requirements.txt
@@ -0,0 +1,7 @@
+kokoro>=0.1.0
+soundfile
+numpy
+torch
+PyYAML
+littlefs-python
+intelhex
diff --git a/firmware/program.bat b/firmware/program.bat
deleted file mode 100644
index 2a210a8..0000000
--- a/firmware/program.bat
+++ /dev/null
@@ -1 +0,0 @@
-nrfutil device --x-ext-mem-config-file buzzy.json program --firmware \\10.0.10.30\edi\nrf_playground\lasertag\firmware\tools\littlefs_generator\lfs_external_flash.hex --options verify=VERIFY_READ,reset=RESET_SYSTEM
\ No newline at end of file