diff --git a/.gitignore b/.gitignore
index 0cf5cb4..33b6c8d 100644
--- a/.gitignore
+++ b/.gitignore
@@ -10,3 +10,4 @@ setup/out-*.reg
 *.wav
 setup/microslopsam.env
 third_party/flextalk/extracted/
+__pycache__/
diff --git a/README.md b/README.md
index f2a74fc..3389bc1 100644
--- a/README.md
+++ b/README.md
@@ -73,8 +73,8 @@ Options: `-v` voice, `-t` text, `-o` output WAV, `-p` pitch, `-s` speed, `-V` vo
     export VESKSLOP_TTS_CMD="$PWD/setup/veskslop-tts.sh {out} {text_file}"
     export VESKSLOP_TTS_VOICES="$(setup/veskslop-tts.sh --voices)"
 
-The wrapper maps voice IDs (`sam`, `mary`, `mike`, `keith-bell`, `kerry-watson`, plus the
-MSTTS variants), reads `VESKSLOP_TTS_VOICE`, `VESKSLOP_TTS_SPEED` (multiplier),
+The wrapper maps voice IDs (`sam`, `mary`, `mike`, `keith-bell`, `kerry-watson`, `vox`,
+`hev`, plus the MSTTS variants), reads `VESKSLOP_TTS_VOICE`, `VESKSLOP_TTS_SPEED` (multiplier),
 `VESKSLOP_TTS_PITCH` (semitones) and `VESKSLOP_TTS_EXPRESSION` (0-100, Flextalk voices
 only), starts Xvfb when `DISPLAY` is unset, and writes a 48 kHz mono s16le WAV trimmed,
 faded and normalized to about -16 LUFS (-3 dBFS true peak).
@@ -95,10 +95,17 @@ so the first word is not gated; tune with `VESKSLOP_TTS_HUM_MS` (0 = off),
 
 **Half-Life presets** (Keith Bell): `VESKSLOP_TTS_PRESET=vox` applies the Valve VOX
 profile (pitch 0, rate 15, head size 34, lowering 50) and enables VOX prosody;
-`VESKSLOP_TTS_PRESET=hev` is a starting point for the HEV suit voice. The VOX values were
+`VESKSLOP_TTS_PRESET=hev` is a starting point for the HEV suit voice. The `vox` and `hev`
+voice IDs select these directly. The VOX values were
 tuned against the game's word assets with `tools/vox_compare.py`; details in
 `docs/TTS-DESIGN.md`.
 
+`VESKSLOP_TTS_CRUNCH` (`hl90`, `hl90hq`, `telephone`, `adpcm`, `crush`) adds 1990s lossy
+character to the output; `VESKSLOP_TTS_CRUNCH_BITS`/`_RATE` override its bit depth and
+sample rate. `VESKSLOP_TTS_STREAM=1` makes word mode write the WAV as it goes (no
+measurement pass; `VESKSLOP_TTS_STREAM_GAIN_DB` sets the gain). Echo timing and level are
+`VESKSLOP_TTS_ECHO_DELAY_MS` and `VESKSLOP_TTS_ECHO_LEVEL`.
+
 ## Layout
 
 - `host/sapi4host.c` — host: voice select, pitch/speed, text -> WAV
@@ -109,6 +116,7 @@ tuned against the game's word assets with `tools/vox_compare.py`; details in
 - `setup/inf2reg.py` — INF AddReg -> `.reg` converter
 - `setup/run.sh` — run helper (prefix + Xvfb)
 - `setup/veskslop-tts.sh` — Veskslop command-provider wrapper (post-processing included)
+- `setup/test-wrapper.sh` — wrapper regression test (stubbed Wine/host, no prefix needed)
 - `setup/vox_parser.py` — NetTTS-style VOX prosody encoder (Flextalk voices)
 - `tools/vox_compare.py` — reference-vs-synthesis metrics for preset tuning
 - `docs/NOTES.md` — registry/CLSID facts, runtime RE notes, gotchas
diff --git a/docs/NOTES.md b/docs/NOTES.md
index 37ae237..a7be815 100644
--- a/docs/NOTES.md
+++ b/docs/NOTES.md
@@ -127,8 +127,8 @@ without them.
   the Python port; outputs are identical on the 31-line corpus. Re-check after any
   parser change while the clone is available.
 - VOX mode is Flextalk-only. The wrapper enables it with `VESKSLOP_TTS_VOX=1`
-  (`VESKSLOP_TTS_VOX_CLEAN=1` drops the stutter wrapper) and it takes precedence
-  over `VESKSLOP_TTS_WORDS`.
+  (`VESKSLOP_TTS_VOX_CLEAN=1` drops the stutter wrapper); word mode
+  (`VESKSLOP_TTS_WORDS=1`) wins and skips VOX prosody.
 - Half-Life presets (`VESKSLOP_TTS_PRESET=vox|hev`) target Keith Bell only. VOX starts
   from NetTTS `nettts_flextalk_vox_profile.reg` (pitch 3, rate 17, head 52, lowering 40)
   and is tuned against the game's word assets from `suchipi/half-life-vox-console`:
diff --git a/docs/TTS-DESIGN.md b/docs/TTS-DESIGN.md
index 5a6c6a6..292ad62 100644
--- a/docs/TTS-DESIGN.md
+++ b/docs/TTS-DESIGN.md
@@ -254,6 +254,9 @@ record what the engine needs to interpret, add, or hide._
 - Assigned stable IDs: `sam`, `mary`, `mike`, `keith-bell`, `kerry-watson`; variants get
   `mike-telephone`, `mike-hall`, `mary-stadium`, `mike-space`, and so on. Unknown IDs fall
   back to `sam`.
+- `vox` (VOX (Half-Life)) and `hev` (HEV suit) are Keith Bell with a fixed preset profile;
+  selecting one implies `VESKSLOP_TTS_PRESET` (`vox` also enables VOX prosody). Their
+  `defaults=` metadata follows the active preset so the UI snap points match.
 - `/voices` (HTTP mode) reports `language: "en"`, gender, and the speed/pitch ranges below.
 
 ### Speed and pitch mapping (measured 2026-09-15)
@@ -372,15 +375,16 @@ Requirements and scope:
   The host passes `TTSDATAFLAG_TAGGED`, so the engine honors them. Other voices ignore
   `VESKSLOP_TTS_VOX`.
 - Needs `python3` at runtime; the wrapper fails loudly when it is missing.
-- Mutually exclusive with `VESKSLOP_TTS_WORDS` (VOX wins).
+- Combines with `VESKSLOP_TTS_WORDS`: word mode wins and VOX prosody is skipped.
 - Measured on the dev host for "Now approaching Sector C test lab. Please stand clear
   of the test chamber.": plain 4.70 s with 1 pause, VOX 7.81 s with 8 pauses.
 
 Other NetTTS features and their relevance:
 
-- Inline `/rate N`, `/pitch N`, `[[pause N]]`: not ported. Veskslop has speed/pitch
-  controls and the profile window; `[[pause N]]` could be added as a wrapper-level
-  text splitter later (NetTTS supports it only outside VOX mode).
+- Inline `/rate N`, `/pitch N` are handled by the wrapper at the start of a line
+  (percent, or a multiplier at or below 2.0); they override the env sliders for that
+  line and are not spoken. `[[pause N]]` is not ported; it could be added as a
+  wrapper-level text splitter later.
 - TCP server with `START`/`STOP` status sockets: not needed; Veskslop owns playback
   and knows when a line ends.
 - Winetricks prefix automation (`winxp`, `vcrun6`, `mfc42`, `riched20`): not needed;
@@ -702,10 +706,16 @@ Inputs: `{out}`, `{text_file}`, optional positional voice/speed/pitch/expression
 | --- | --- |
 | `VESKSLOP_TTS_PROFILE` | `pitch:27:...`/`key=value,...` overrides; the wrapper writes the full profile (defaults + overrides) to the Wine registry with `wine reg import` before every synthesis. |
 | `VESKSLOP_TTS_HIGHPASS` | High-pass Hz, 40-400 (default 40). |
-| `VESKSLOP_TTS_ECHO` | Mono echo 0-100 (`aecho`), default 0. |
+| `VESKSLOP_TTS_ECHO` | Mono echo 0-100 (`aecho`), default 0. The echo is mixed under the dry voice, so the voice level does not change. |
+| `VESKSLOP_TTS_ECHO_DELAY_MS` / `VESKSLOP_TTS_ECHO_LEVEL` | Echo delay 10-400 ms (default 60) and repeat level 0-0.95 relative to the dry voice (default 0.85). |
+| `VESKSLOP_TTS_CRUNCH` | `off` (default) or `hl90`, `hl90hq`, `telephone`, `adpcm`, `crush`: 1990s lossy character, applied last. |
+| `VESKSLOP_TTS_CRUNCH_BITS` / `VESKSLOP_TTS_CRUNCH_RATE` | Bit depth 1-16 and sample rate 4000-48000 overrides for the crunch stage. |
+| `VESKSLOP_TTS_STREAM` / `VESKSLOP_TTS_STREAM_GAIN_DB` | `1` = word mode writes the WAV as it goes (no measurement pass); fixed gain in dB for that path. |
 | `VESKSLOP_TTS_WORDS` / `VESKSLOP_TTS_WORD_GAP_MS` | Word-by-word synthesis with silence gaps (Flextalk only). |
 
-Placeholders `{profile}`, `{highpass}`, `{echo}`, `{words}`, `{word_gap_ms}` mirror the env vars.
+Placeholders `{profile}`, `{highpass}`, `{echo}`, `{echo_delay_ms}`, `{echo_level}`,
+`{words}`, `{word_gap_ms}`, `{crunch}`, `{crunch_bits}`, `{crunch_rate}`, `{stream}`,
+`{vox}`, `{vox_clean}`, `{preset}` mirror the env vars.
 `wine regedit` fails headlessly but returns 0: use `wine reg import`. A missing profile value
 must always write the engine default so no stale registry state survives.
 
diff --git a/setup/test-wrapper.sh b/setup/test-wrapper.sh
index a218309..0144efd 100755
--- a/setup/test-wrapper.sh
+++ b/setup/test-wrapper.sh
@@ -94,8 +94,8 @@ check "vox preset implies VOX parser" "1" "$(grep -c 'wH1' "$STUB_LOG")"
 # The preset writes the full audible profile to both hives.
 VESKSLOP_TTS_VOICE=keith-bell VESKSLOP_TTS_PRESET=vox run_wrapper
 check "preset writes both hives" "2" "$(grep -c 'Speakers\\{11260610' "$STUB_REG")"
-check "preset pitch (both hives)" "2" "$(grep -c '"Pitch"=dword:00000003' "$STUB_REG")"
-check "preset rate (both hives)" "2" "$(grep -c '"Speaking Rate"=dword:00000011' "$STUB_REG")"
+check "preset pitch (both hives)" "2" "$(grep -c '"Pitch"=dword:00000000' "$STUB_REG")"
+check "preset rate (both hives)" "2" "$(grep -c '"Speaking Rate"=dword:0000000f' "$STUB_REG")"
 check "preset gender and line mode" "4" "$(grep -cE '"Gender"=dword:00000002|"Line Mode"=dword:00000001' "$STUB_REG")"
 check "preset default speaker" "2" "$(grep -c '"Default Speaker"' "$STUB_REG")"
 
@@ -111,7 +111,7 @@ check "word mode with preset synthesizes every word" "3" "$(grep -c '^ARGV:' "$S
 check "word mode skips the VOX parser" "0" "$(grep -c 'wH1' "$STUB_LOG")"
 check "word mode tags every word" "3" \
     "$(sed -n 's/.*\[-t\] \[\(.*\)\] \[-o\].*/\1/p' "$STUB_LOG" | grep -c '^\\!R0.80 ')"
-check "word mode keeps the preset profile" "2" "$(grep -c '"Pitch"=dword:00000003' "$STUB_REG")"
+check "word mode keeps the preset profile" "2" "$(grep -c '"Pitch"=dword:00000000' "$STUB_REG")"
 check "word mode emits no attributes" "0" "$(attr_count)"
 
 # The vox preset implies VOX when the variable is unset, but word mode still wins.
@@ -139,7 +139,7 @@ check "stream has a gap per word break" "2" \
 
 # The Half-Life voices carry their profile preset.
 VESKSLOP_TTS_VOICE=vox VESKSLOP_TTS_VOX=1 run_wrapper
-check "vox voice writes the VOX profile (both hives)" "2" "$(grep -c '"Pitch"=dword:00000003' "$STUB_REG")"
+check "vox voice writes the VOX profile (both hives)" "2" "$(grep -c '"Pitch"=dword:00000000' "$STUB_REG")"
 check "vox voice runs VOX prosody" "1" "$(grep -c 'wH1' "$STUB_LOG")"
 VESKSLOP_TTS_VOICE=hev run_wrapper
 check "hev voice writes the HEV profile (both hives)" "2" "$(grep -c '"Pitch"=dword:0000000f' "$STUB_REG")"
diff --git a/setup/veskslop-tts.sh b/setup/veskslop-tts.sh
index 137c9ce..ee6737a 100755
--- a/setup/veskslop-tts.sh
+++ b/setup/veskslop-tts.sh
@@ -16,13 +16,24 @@
 #                                             prominences, head_size, lowering
 #                   VESKSLOP_TTS_HIGHPASS     high-pass Hz (40-400, default 40)
 #                   VESKSLOP_TTS_ECHO         mono echo amount 0-100 (default 0)
-#                   VESKSLOP_TTS_WORDS        1 = synthesize each word, merge (Flextalk)
+#                   VESKSLOP_TTS_ECHO_DELAY_MS  echo delay 10-400 (default 60)
+#                   VESKSLOP_TTS_ECHO_LEVEL   echo repeat level 0-0.95, relative to the
+#                                             dry voice (default 0.85)
+#                   VESKSLOP_TTS_CRUNCH       off | hl90 | hl90hq | telephone | adpcm | crush
+#                                             (1990s lossy character, applied last)
+#                   VESKSLOP_TTS_CRUNCH_BITS  1-16, overrides the preset bit depth
+#                   VESKSLOP_TTS_CRUNCH_RATE  4000-48000 Hz, overrides the preset rate
+#                   VESKSLOP_TTS_STREAM       1 = word mode writes the WAV as it goes
+#                   VESKSLOP_TTS_STREAM_GAIN_DB  fixed gain for streaming (no measurement)
+#                   VESKSLOP_TTS_WORDS        1 = synthesize each word, merge (Flextalk);
+#                                             combines with a preset and wins over VOX prosody
 #                   VESKSLOP_TTS_WORD_GAP_MS  silence between words in ms (0-500)
 #                   VESKSLOP_TTS_VOX          1 = VOX prosody, NetTTS-style (Flextalk)
 #                   VESKSLOP_TTS_VOX_CLEAN    1 = VOX without the \!wH1/\!wH0 stutter
 #                   VESKSLOP_TTS_HUM_MS       leading low hum in ms, 0 = off (default 80)
 #                   VESKSLOP_TTS_HUM_HZ       hum frequency, 30-200 (default 75)
 #                   VESKSLOP_TTS_HUM_DB       hum peak level, -80..-6 (default -48)
+#                   voice `vox`/`hev`         Half-Life profiles as first-class voices
 #                   VESKSLOP_TTS_PRESET       vox | hev (Half-Life profiles, Keith Bell;
 #                                             vox also enables VOX prosody)
 #
@@ -33,11 +44,21 @@ HERE=$(CDPATH= cd -- "$(dirname -- "$0")" && pwd)
 . "$HERE/config.sh"
 ROOT=$MICROSLOP_ROOT
 
+# The Keith Bell defaults follow the active preset so the UI's "default" snap points
+# match the values the preset actually writes.
+VOX_DEFAULTS="pitch:0:rate:15:prominences:22:head_size:34:lowering:50:expression:59"
+HEV_DEFAULTS="pitch:15:rate:24:prominences:22:head_size:52:lowering:40:expression:59"
+case "${VESKSLOP_TTS_PRESET:-}" in
+    vox) KB_DEFAULTS="$VOX_DEFAULTS" ;;
+    hev) KB_DEFAULTS="$HEV_DEFAULTS" ;;
+    *)   KB_DEFAULTS="pitch:27:rate:33:prominences:17:head_size:46:lowering:40:expression:59" ;;
+esac
+
 # VESKSLOP_TTS_VOICES entries: id|Name|gender|language[|flags]
 # `expression=NN` marks a Flextalk voice with an Expression profile parameter and
 # its registry default. Expression is applied at voice load, so the wrapper writes
 # it into the Wine registry before synthesis.
-VOICES="sam|Sam|male|en,mary|Mary|female|en,mike|Mike|male|en,keith-bell|Keith Bell|male|en|expression=59+defaults=pitch:27:rate:33:prominences:17:head_size:46:lowering:40:expression:59,kerry-watson|Kerry Watson|female|en|expression=100+defaults=pitch:65:rate:32:prominences:17:head_size:45:lowering:40:expression:100,mike-telephone|Mike (for Telephone)|male|en,mary-telephone|Mary (for Telephone)|female|en,mike-hall|Mike in Hall|male|en,mike-stadium|Mike in Stadium|male|en,mike-space|Mike in Space|male|en,mary-hall|Mary in Hall|female|en,mary-stadium|Mary in Stadium|female|en,mary-space|Mary in Space|female|en"
+VOICES="sam|Sam|male|en,mary|Mary|female|en,mike|Mike|male|en,keith-bell|Keith Bell|male|en|expression=59+defaults=$KB_DEFAULTS,vox|VOX (Half-Life)|male|en|expression=59+defaults=$VOX_DEFAULTS,hev|HEV suit|male|en|expression=59+defaults=$HEV_DEFAULTS,kerry-watson|Kerry Watson|female|en|expression=100+defaults=pitch:65:rate:32:prominences:17:head_size:45:lowering:40:expression:100,mike-telephone|Mike (for Telephone)|male|en,mary-telephone|Mary (for Telephone)|female|en,mike-hall|Mike in Hall|male|en,mike-stadium|Mike in Stadium|male|en,mike-space|Mike in Space|male|en,mary-hall|Mary in Hall|female|en,mary-stadium|Mary in Stadium|female|en,mary-space|Mary in Space|female|en"
 
 if [ "${1:-}" = "--voices" ]; then
     echo "$VOICES"
@@ -53,11 +74,25 @@ EXPRESSION=${VESKSLOP_TTS_EXPRESSION:-${6:-}}
 PROFILE=${VESKSLOP_TTS_PROFILE:-}
 HIGHPASS=${VESKSLOP_TTS_HIGHPASS:-40}
 ECHO=${VESKSLOP_TTS_ECHO:-0}
+ECHO_DELAY=${VESKSLOP_TTS_ECHO_DELAY_MS:-60}
+ECHO_LEVEL=${VESKSLOP_TTS_ECHO_LEVEL:-0.85}
+CRUNCH=${VESKSLOP_TTS_CRUNCH:-off}
+CRUNCH_BITS=${VESKSLOP_TTS_CRUNCH_BITS:-}
+CRUNCH_RATE=${VESKSLOP_TTS_CRUNCH_RATE:-}
+STREAM=${VESKSLOP_TTS_STREAM:-0}
+STREAM_GAIN=${VESKSLOP_TTS_STREAM_GAIN_DB:-0}
 WORDS=${VESKSLOP_TTS_WORDS:-0}
 WORD_GAP=${VESKSLOP_TTS_WORD_GAP_MS:-120}
 VOX=${VESKSLOP_TTS_VOX:-0}
 VOX_CLEAN=${VESKSLOP_TTS_VOX_CLEAN:-0}
 PRESET=${VESKSLOP_TTS_PRESET:-}
+
+# The Half-Life voices are Keith Bell with a fixed profile preset; selecting one
+# stands in for VESKSLOP_TTS_PRESET.
+case "$VOICE" in
+    vox) [ -n "$PRESET" ] || PRESET=vox ;;
+    hev) [ -n "$PRESET" ] || PRESET=hev ;;
+esac
 # The VOX preset is the Half-Life announcement voice: it implies VOX prosody
 # unless the caller explicitly disables it.
 if [ "$PRESET" = "vox" ] && [ -z "${VESKSLOP_TTS_VOX:-}" ]; then
@@ -72,6 +107,8 @@ case "$VOICE" in
     mary)           ENGVOICE="Mary" ;;
     mike)           ENGVOICE="Mike" ;;
     keith-bell)     ENGVOICE="Keith Bell" ;;
+    vox)            ENGVOICE="Keith Bell" ;;
+    hev)            ENGVOICE="Keith Bell" ;;
     kerry-watson)   ENGVOICE="Kerry Watson" ;;
     mike-telephone) ENGVOICE="Mike (for Telephone)" ;;
     mary-telephone) ENGVOICE="Mary (for Telephone)" ;;
@@ -350,9 +387,11 @@ fi
 
 # SAPI4 is ANSI/CP1252. VOX mode (Flextalk only) first transforms the text into
 # vendor tags (\!wH1/\!wH0, \!br, \!sfNN) with setup/vox_parser.py; the parser
-# works on UTF-8, so it runs before the CP1252 transliteration.
+# works on UTF-8, so it runs before the CP1252 transliteration. Word mode builds
+# its own cadence from single-word instances, so the parser is skipped there: VOX
+# tags cannot survive a word split.
 TEXT_SRC="$TEXT_IN"
-if [ "$VOX" = "1" ] && [ -n "$SPEAKER_GUID" ]; then
+if [ "$VOX" = "1" ] && [ -n "$SPEAKER_GUID" ] && [ "$WORDS" != "1" ]; then
     if command -v "$MICROSLOP_PYTHON" >/dev/null 2>&1 && [ -f "$MICROSLOP_ROOT/setup/vox_parser.py" ]; then
         if [ "$VOX_CLEAN" = "1" ]; then
             "$MICROSLOP_PYTHON" "$MICROSLOP_ROOT/setup/vox_parser.py" --clean < "$TEXT_IN" > "$TMP/vox.txt"
@@ -399,10 +438,19 @@ try_run_host() {
     exit 1
 }
 
+# Streaming word mode writes each word as it is synthesized (see below), after the
+# filter chain exists, so the classic all-at-once path is skipped for it.
+STREAM_WORDS=0
+if [ "$STREAM" = "1" ] && [ "$WORDS" = "1" ] && [ -n "$SPEAKER_GUID" ]; then
+    STREAM_WORDS=1
+fi
+
+if [ "$STREAM_WORDS" != "1" ]; then
 # Word mode (Flextalk): synthesize each word and merge with silence gaps, which
 # gives the staccato radio cadence the Half-Life VOX/HEV voices are known for.
-# VOX mode takes precedence: it produces its own cadence with vendor tags.
-if [ "$WORDS" = "1" ] && [ -n "$SPEAKER_GUID" ] && [ "$VOX" != "1" ]; then
+# It combines with a preset (the profile is already written) and takes precedence
+# over VOX prosody, which cannot survive a word split.
+if [ "$WORDS" = "1" ] && [ -n "$SPEAKER_GUID" ]; then
     GAP=$WORD_GAP
     case "$GAP" in *[!0-9]*|"") GAP=120 ;; esac
     [ "$GAP" -gt 500 ] && GAP=500
@@ -432,6 +480,7 @@ else
     try_run_host "$RAW" "$TAG_PREFIX$(cat "$TXT")"
 fi
 [ -s "$RAW" ] || { echo "engine produced no audio" >&2; exit 1; }
+fi
 
 # Filter chain: adjustable high-pass, silence trim, fades, optional mono echo.
 case "$HIGHPASS" in *[!0-9]*|"") HIGHPASS=40 ;; esac
@@ -441,17 +490,65 @@ case "$HIGHPASS" in *[!0-9]*|"") HIGHPASS=40 ;; esac
 case "$ECHO" in *[!0-9]*|"") ECHO=0 ;; esac
 [ "$ECHO" -gt 100 ] && ECHO=100
 
+case "$ECHO_DELAY" in *[!0-9]*|"") ECHO_DELAY=60 ;; esac
+[ "$ECHO_DELAY" -lt 10 ] && ECHO_DELAY=10
+[ "$ECHO_DELAY" -gt 400 ] && ECHO_DELAY=400
+case "$ECHO_LEVEL" in *[!0-9.]*|""|".") ECHO_LEVEL=0.85 ;; esac
+ECHO_LEVEL=$(awk -v v="$ECHO_LEVEL" 'BEGIN{ if (v<0) v=0; if (v>0.95) v=0.95; printf "%.2f", v }')
+
 SIL="silenceremove=start_periods=1:start_duration=0.05:start_silence=0.03:start_threshold=-50dB"
 PRE="highpass=f=$HIGHPASS,$SIL,afade=t=in:st=0:d=0.008,areverse,$SIL,afade=t=in:st=0:d=0.008,areverse"
+# aecho's in_gain/out_gain scale the dry signal as well, so a low ECHO_LEVEL used to
+# quiet the whole line (and the streaming path, which skips the loudness measurement,
+# had nothing to make it up). Keep both gains at unity and let the repeats' decay carry
+# the level; the limiter absorbs the summed peaks.
 if [ "$ECHO" -gt 0 ]; then
-    DECAY=$(awk -v e="$ECHO" 'BEGIN{printf "%.2f", 0.2 + e * 0.006}')
-    PRE="$PRE,aecho=0.8:0.85:60:${DECAY}"
+    DECAY=$(awk -v e="$ECHO" -v l="$ECHO_LEVEL" 'BEGIN{
+        d = (0.2 + e * 0.006) * (l / 0.85);
+        if (d < 0.02) d = 0.02;
+        if (d > 0.95) d = 0.95;
+        printf "%.2f", d
+    }')
+    PRE="$PRE,aecho=1:1:${ECHO_DELAY}:${DECAY},alimiter=limit=0.98:level=disabled"
+fi
+
+# Compression stage ("crunch"): the lossy 1990s codec character, built as its own
+# stage that runs after the rest of the chain so it colours the echo tails too. The
+# signal is quantized at a low rate / bit depth and comes back to 48 kHz so the
+# daemon's output contract holds.
+CRUNCH_FILTERS=""
+case "$CRUNCH" in
+    hl90|vox90)      CRUNCH_BAND="highpass=f=200,lowpass=f=5000";  CRUNCH_RATE=${CRUNCH_RATE:-11025}; CRUNCH_BITS=${CRUNCH_BITS:-8} ;;
+    hl90hq)          CRUNCH_BAND="highpass=f=200,lowpass=f=7000";  CRUNCH_RATE=${CRUNCH_RATE:-22050}; CRUNCH_BITS=${CRUNCH_BITS:-8} ;;
+    telephone|phone) CRUNCH_BAND="highpass=f=300,lowpass=f=3400";  CRUNCH_RATE=${CRUNCH_RATE:-8000};  CRUNCH_BITS=${CRUNCH_BITS:-8} ;;
+    adpcm)           CRUNCH_BAND="highpass=f=200,lowpass=f=5000";  CRUNCH_RATE=${CRUNCH_RATE:-11025}; CRUNCH_BITS=${CRUNCH_BITS:-4} ;;
+    crush|custom)    CRUNCH_BAND="${VESKSLOP_TTS_CRUNCH_BAND:-}";  CRUNCH_RATE=${CRUNCH_RATE:-22050}; CRUNCH_BITS=${CRUNCH_BITS:-6} ;;
+    off|"") ;;
+    *) echo "unknown crunch preset: $CRUNCH" >&2; exit 2 ;;
+esac
+if [ "$CRUNCH" != "off" ] && [ -n "$CRUNCH" ]; then
+    case "$CRUNCH_RATE" in *[!0-9]*|"") CRUNCH_RATE=22050 ;; esac
+    [ "$CRUNCH_RATE" -lt 4000 ] && CRUNCH_RATE=4000
+    [ "$CRUNCH_RATE" -gt 48000 ] && CRUNCH_RATE=48000
+    case "$CRUNCH_BITS" in *[!0-9]*|"") CRUNCH_BITS=8 ;; esac
+    [ "$CRUNCH_BITS" -lt 1 ] && CRUNCH_BITS=1
+    [ "$CRUNCH_BITS" -gt 16 ] && CRUNCH_BITS=16
+    [ -n "$CRUNCH_BAND" ] && CRUNCH_FILTERS="$CRUNCH_BAND,"
+    CRUNCH_FILTERS="${CRUNCH_FILTERS}aresample=$CRUNCH_RATE"
+    if [ "$CRUNCH_BITS" = "8" ]; then
+        CRUNCH_FILTERS="$CRUNCH_FILTERS,aformat=sample_fmts=u8"
+    else
+        CRUNCH_FILTERS="$CRUNCH_FILTERS,acrusher=bits=$CRUNCH_BITS:mode=log:mix=1"
+    fi
+    CRUNCH_FILTERS="$CRUNCH_FILTERS,aresample=48000,aformat=sample_fmts=s16"
 fi
 
 # Fixed-gain normalization: single-pass loudnorm misbehaves on short lines
 # (LRA=0 causes heavy attenuation), so measure EBU R128 integrated loudness and
 # true peak, then apply one gain clamped by the -3 dBFS peak target.
-M=$("$MICROSLOP_FFMPEG" -hide_banner -nostats -i "$RAW" -af "$PRE,ebur128=peak=true" \
+MEASURE_CHAIN="$PRE"
+[ -n "$CRUNCH_FILTERS" ] && MEASURE_CHAIN="$PRE,$CRUNCH_FILTERS"
+M=$("$MICROSLOP_FFMPEG" -hide_banner -nostats -i "$RAW" -af "$MEASURE_CHAIN,ebur128=peak=true" \
         -f null - 2>&1 || true)
 I=$(printf '%s' "$M" | sed -n 's/ *I: *\(-\{0,1\}[0-9.]*\) *LUFS.*/\1/p' | tail -1)
 TP=$(printf '%s' "$M" | sed -n 's/ *Peak: *\(-\{0,1\}[0-9.]*\) *dBFS.*/\1/p' | tail -1)
@@ -468,6 +565,71 @@ else
     GAIN=0
 fi
 
+# Streaming word mode: write a WAV header with unknown sizes, then append each
+# word's PCM as it is ready, so a player can start on the first word instead of
+# waiting for the whole line. The loudness measurement is skipped; set
+# VESKSLOP_TTS_STREAM_GAIN_DB if the engine levels need a nudge.
+if [ "$STREAM_WORDS" = "1" ]; then
+    SGAIN=""
+    case "$STREAM_GAIN" in
+        ''|*[!0-9.-]*|'.'|'-') ;;
+        *) SGAIN=",volume=${STREAM_GAIN}dB" ;;
+    esac
+
+    # RIFF/????/WAVE, fmt 16 bytes, PCM mono 48000 Hz s16le, data/????
+    printf 'RIFF\377\377\377\377WAVEfmt \020\000\000\000\001\000\001\000\200\273\000\000\000w\001\000\002\000\020\000data\377\377\377\377' > "$OUT" \
+        || { echo "could not write $OUT" >&2; exit 1; }
+
+    set -f
+    # shellcheck disable=SC2046
+    set -- $(cat "$TXT")
+    set +f
+    WORD_COUNT=$#
+    INDEX=0
+    GAP=$WORD_GAP
+    case "$GAP" in *[!0-9]*|"") GAP=120 ;; esac
+    [ "$GAP" -gt 500 ] && GAP=500
+
+    # Leading hum, written once before the first word (same settings as the classic path).
+    if [ "$HUM_MS" -gt 0 ]; then
+        HUM_S=$(awk -v ms="$HUM_MS" 'BEGIN{printf "%.3f", ms/1000}')
+        HUM_HZ_C=$(awk -v v="$HUM_HZ" 'BEGIN{ if (v<30) v=30; if (v>200) v=200; printf "%.1f", v }')
+        HUM_DB_C=$(awk -v v="$HUM_DB" 'BEGIN{ if (v>-6) v=-6; if (v<-80) v=-80; printf "%.1f", v }')
+        HUM_AMP_C=$(awk -v db="$HUM_DB_C" 'BEGIN{printf "%.6f", 10^(db/20)}')
+        if "$MICROSLOP_FFMPEG" -hide_banner -loglevel error -f lavfi \
+                -i "aevalsrc=exprs='$HUM_AMP_C*sin(2*PI*$HUM_HZ_C*t)':s=48000:d=$HUM_S" \
+                -f s16le -ac 1 -ar 48000 "$TMP/hum.raw"; then
+            cat "$TMP/hum.raw" >> "$OUT"
+        fi
+    fi
+
+    for word in "$@"; do
+        INDEX=$((INDEX + 1))
+        try_run_host "$TMP/sw$INDEX.wav" "${TAG_PREFIX}${word}"
+        if ! "$MICROSLOP_FFMPEG" -hide_banner -loglevel error -y -i "$TMP/sw$INDEX.wav" \
+                -af "$PRE$SGAIN${CRUNCH_FILTERS:+,$CRUNCH_FILTERS}" \
+                -f s16le -ac 1 -ar 48000 "$TMP/sw$INDEX.raw"; then
+            echo "streaming post-processing failed" >&2
+            exit 1
+        fi
+        cat "$TMP/sw$INDEX.raw" >> "$OUT" 2>/dev/null || { echo "stream write failed" >&2; exit 1; }
+
+        # The player waits for this marker so it does not start on the bare header.
+        if [ "$INDEX" = "1" ]; then : > "$OUT.ready"; fi
+
+        if [ "$GAP" -gt 0 ] && [ "$INDEX" -lt "$WORD_COUNT" ]; then
+            if "$MICROSLOP_FFMPEG" -hide_banner -loglevel error -y -f lavfi -i "anullsrc=r=48000:cl=mono" \
+                    -t "$(awk -v g="$GAP" 'BEGIN{printf "%.3f", g/1000}')" \
+                    -f s16le -ac 1 -ar 48000 "$TMP/gap.raw"; then
+                cat "$TMP/gap.raw" >> "$OUT"
+            fi
+        fi
+    done
+
+    [ "$INDEX" -gt 0 ] || { echo "no words to stream" >&2; exit 1; }
+    exit 0
+fi
+
 TMPOUT="$TMP/out.wav"
 # Leading hum: a quiet low tone in the pre-speech padding, to prime Discord's
 # audio pipeline so the first syllable is not gated. Level/frequency/duration
@@ -488,12 +650,12 @@ if [ "$HUM_MS" -gt 0 ]; then
     if ! "$MICROSLOP_FFMPEG" -hide_banner -nostats -y \
             -f lavfi -i "aevalsrc=exprs='$HUM_AMP*sin(2*PI*$HUM_HZ*t)':s=48000:d=$HUM_S" \
             -i "$RAW" \
-            -filter_complex "[0:a]afade=t=in:st=0:d=$HUM_FADE,afade=t=out:st=$HUM_FADE_OUT:d=$HUM_FADE,aformat=sample_fmts=s16:channel_layouts=mono:sample_rates=48000[h];[1:a]$PRE,volume=${GAIN}dB,aresample=48000,aformat=sample_fmts=s16:channel_layouts=mono[a];[h][a]concat=n=2:v=0:a=1[out]" \
+            -filter_complex "[0:a]afade=t=in:st=0:d=$HUM_FADE,afade=t=out:st=$HUM_FADE_OUT:d=$HUM_FADE,aformat=sample_fmts=s16:channel_layouts=mono:sample_rates=48000[h];[1:a]$PRE,volume=${GAIN}dB,${CRUNCH_FILTERS:+$CRUNCH_FILTERS,}aresample=48000,aformat=sample_fmts=s16:channel_layouts=mono[a];[h][a]concat=n=2:v=0:a=1[out]" \
             -map "[out]" -c:a pcm_s16le "$TMPOUT" >/dev/null 2>"$TMP/ff"; then
         head -c 400 "$TMP/ff" >&2
         exit 1
     fi
-elif ! "$MICROSLOP_FFMPEG" -hide_banner -nostats -y -i "$RAW" -af "$PRE,volume=${GAIN}dB" \
+elif ! "$MICROSLOP_FFMPEG" -hide_banner -nostats -y -i "$RAW" -af "$PRE,volume=${GAIN}dB,${CRUNCH_FILTERS:+$CRUNCH_FILTERS,}aresample=48000,aformat=sample_fmts=s16" \
         -ar 48000 -ac 1 -c:a pcm_s16le "$TMPOUT" >/dev/null 2>"$TMP/ff"; then
     head -c 400 "$TMP/ff" >&2
     exit 1