sessionvoice 0.3.1__tar.gz → 0.3.2__tar.gz

{sessionvoice-0.3.1 → sessionvoice-0.3.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sessionvoice
-Version: 0.3.1
+Version: 0.3.2
 Summary: One listening engine for coding agents and the Delta bot: Apple / Nemotron / whisper -> one event stream -> VAD -> turns
 Author-email: Mike Bailey <mike@sessionkit.ai>
 License-Expression: AGPL-3.0-or-later
@@ -133,10 +133,15 @@ pair of ears on a Mac (depth in [docs/architecture.md](docs/architecture.md)):
   package or a checkout; a bare install never does.
 - **Turn mail** (`keeper/ears-mail.py`, launchd `dev.sessionkit.voice.ears-mail`):
   mails each of the user's finished turns To and From
-  `heard+<device>@<host>`, with the `Ears-*` headers of the pager genie's
-  contract (sessionmail `openspec/changes/pager-genie`). A session hears a
-  device by subscribing to that address. The box is `heard` (`--sender`),
-  because `ears` is this project's own box.
+  `heard+room@<host>`, whatever input device the capture is following, with
+  the `Ears-*` headers of the pager genie's contract (sessionmail
+  `openspec/changes/pager-genie`). The device is named by the `Ears-Device`
+  header. A session hears the room, on every device, by subscribing to that
+  one address. While `--device-copy` is on (the default), the same send (one
+  message, one Message-ID) also goes Cc `heard+<device>@<host>`, so a
+  session subscribed to a device's address keeps hearing that device. The
+  room's tag is `--room` (`EARS_MAIL_ROOM`, default `room`). The box is
+  `heard` (`--sender`), because `ears` is this project's own box.
   sessionmail trusts only its keeper's From (`_voice.KEEPER`), so the
   plist loads only once that says `heard` too. `--to seats` is the older mail, To
   each seat a turn is for. **Never run `--to seats` and `--to device` for
@@ -239,21 +244,23 @@ Every key, its type and its default:
 - **The heard log:** `<home>/logs/conversations/heard_YYYY-MM-DD.jsonl`,
   one JSON line for each thing the ears heard and the mouth said, where
   `<home>` is the home above. A new file starts each day.
-- **The services' logs:** `~/Library/Logs/ears/` (`EARS_LOG_DIR` moves it).
+- **The services' logs:** `~/.sessionkit/voice/log/` (`EARS_LOG_DIR` moves it).
   `ears-keeper.log` is what the keeper did (devices, restarts, calls, a
   deaf mic); `ears-keeper-capture.log` is the room capture's own output;
   `ears-mail.log` is each turn mail sent or skipped; `ears-mouth.log` is
-  the mouth's player.
+  the mouth's player. Logs written before the move stay where they
+  were, and nothing deletes them. `ears services status` prints the folder.
 
 Follow one with `tail -F`, for example
-`tail -F ~/Library/Logs/ears/ears-keeper.log`, or today's heard log with
+`tail -F ~/.sessionkit/voice/log/ears-keeper.log`, or today's heard log with
 `tail -F <home>/logs/conversations/heard_$(date +%F).jsonl`.
 Every log file and what each line means:
 [docs/logs.md](docs/logs.md).
 
 - **Health:** `ears services status` prints one row per launch agent
-  (`running, pid N`, `loaded, not running` or `not installed`), the
-  keeper's `mode` (`input-watch`, or `polling system_profiler` and why),
+  (`running, pid N`, `loaded, not running` or `not installed`; the
+  `stt` row, the optional `--apple` server, also names what the ears hear
+  through), the keeper's `mode` (`input-watch`, or `polling system_profiler` and why),
   the one that serves speech on `127.0.0.1:8890`, and the home in use.
 
 ## Use
@@ -297,11 +304,17 @@ at `http://127.0.0.1:<port>/`, never at `localhost`. The code is `src/talk/`.
 `ears services install --only talk` keeps it running as a launch agent
 (opt-in; a bare install never does).
 
+In the kit's web page, Talk is a room: the kit's front door runs
+`talk serve --kit` on a unix socket and logs you in, so there is nothing to
+pair and you never run `--kit` by hand
+([docs/talk.md](docs/talk.md#in-the-kits-web-page)).
+
 ## Install
 
 The distribution is `sessionvoice` (`sessionvoice[nemotron]`, `sessionvoice[all]`);
-it installs the `ears`, `ears-serve`, `mouth` and `talk` commands. A bare
-`pip install sessionvoice` carries what `mouth say` needs. The extras are
+it installs the `ears`, `ears-serve`, `mouth`, `talk` and `sessionvoice-web`
+commands. A bare `pip install sessionvoice` carries what `mouth say` needs.
+The extras are
 `capture` (the mic and call ears), `silero` (the Silero VAD), `smart-turn`
 (the Smart Turn gate's ONNX backend), `nemotron` (mlx-audio) and `all`. An
 import that needs one fails naming it, as `pip install 'sessionvoice[EXTRA]'`.
@@ -339,7 +352,7 @@ with its fix, writes nothing and exits 2. A launch agent that fails to load
 also makes it exit 2, once every other service has been tried: one block on
 stderr names each service that did not load, the install log and `ears
 services status`. Every real run appends its output and launchctl's answers,
-timestamped, to `<sessionvoice's home>/logs/install.log`, and prints that path
+timestamped, to `install.log` in the log folder, `~/.sessionkit/voice/log/`, and prints that path
 as its `log` row ([logs](docs/logs.md#the-log-files)). It then:
 
 - first finds the mlx-audio speech server on `:8890`, or installs its own
@@ -390,7 +403,7 @@ as its `log` row ([logs](docs/logs.md#the-log-files)). It then:
   read (no `message`, or one older than 0.2.0), it warns, makes no box,
   checks neither, and goes on.
   On a postfix host it also prints the box to subscribe to:
-  `message subscription add heard+macbook`;
+  `message subscription add heard+room`;
 - `--agent NAME` adds NAME's wake seat (`ears wake add NAME`) and sends the
   keeper's notices to NAME (`EARS_NOTIFY`; `--notify ADDR` sends them
   elsewhere). Without it, and when the wake file names no seat, nobody would
@@ -455,7 +468,7 @@ the agent's next tool call. `EARS_HEARD_HOOK=off` silences it without
 editing the file.
 
 `ears heard hook` is the canonical heard hook. A seat that also gets heard
-mail (it subscribes to `heard+<device>`) gets each finished turn by mail, so
+mail (it subscribes to `heard+room`) gets each finished turn by mail, so
 give it `ears heard hook --partials-only || exit 1` instead: the hook then
 shows only the partial preview of a turn still being spoken and the shown
 events (wake word, barge-in, device change), never the finished turn a

{sessionvoice-0.3.1 → sessionvoice-0.3.2}/README.md

@@ -89,10 +89,15 @@ pair of ears on a Mac (depth in [docs/architecture.md](docs/architecture.md)):
   package or a checkout; a bare install never does.
 - **Turn mail** (`keeper/ears-mail.py`, launchd `dev.sessionkit.voice.ears-mail`):
   mails each of the user's finished turns To and From
-  `heard+<device>@<host>`, with the `Ears-*` headers of the pager genie's
-  contract (sessionmail `openspec/changes/pager-genie`). A session hears a
-  device by subscribing to that address. The box is `heard` (`--sender`),
-  because `ears` is this project's own box.
+  `heard+room@<host>`, whatever input device the capture is following, with
+  the `Ears-*` headers of the pager genie's contract (sessionmail
+  `openspec/changes/pager-genie`). The device is named by the `Ears-Device`
+  header. A session hears the room, on every device, by subscribing to that
+  one address. While `--device-copy` is on (the default), the same send (one
+  message, one Message-ID) also goes Cc `heard+<device>@<host>`, so a
+  session subscribed to a device's address keeps hearing that device. The
+  room's tag is `--room` (`EARS_MAIL_ROOM`, default `room`). The box is
+  `heard` (`--sender`), because `ears` is this project's own box.
   sessionmail trusts only its keeper's From (`_voice.KEEPER`), so the
   plist loads only once that says `heard` too. `--to seats` is the older mail, To
   each seat a turn is for. **Never run `--to seats` and `--to device` for
@@ -195,21 +200,23 @@ Every key, its type and its default:
 - **The heard log:** `<home>/logs/conversations/heard_YYYY-MM-DD.jsonl`,
   one JSON line for each thing the ears heard and the mouth said, where
   `<home>` is the home above. A new file starts each day.
-- **The services' logs:** `~/Library/Logs/ears/` (`EARS_LOG_DIR` moves it).
+- **The services' logs:** `~/.sessionkit/voice/log/` (`EARS_LOG_DIR` moves it).
   `ears-keeper.log` is what the keeper did (devices, restarts, calls, a
   deaf mic); `ears-keeper-capture.log` is the room capture's own output;
   `ears-mail.log` is each turn mail sent or skipped; `ears-mouth.log` is
-  the mouth's player.
+  the mouth's player. Logs written before the move stay where they
+  were, and nothing deletes them. `ears services status` prints the folder.
 
 Follow one with `tail -F`, for example
-`tail -F ~/Library/Logs/ears/ears-keeper.log`, or today's heard log with
+`tail -F ~/.sessionkit/voice/log/ears-keeper.log`, or today's heard log with
 `tail -F <home>/logs/conversations/heard_$(date +%F).jsonl`.
 Every log file and what each line means:
 [docs/logs.md](docs/logs.md).
 
 - **Health:** `ears services status` prints one row per launch agent
-  (`running, pid N`, `loaded, not running` or `not installed`), the
-  keeper's `mode` (`input-watch`, or `polling system_profiler` and why),
+  (`running, pid N`, `loaded, not running` or `not installed`; the
+  `stt` row, the optional `--apple` server, also names what the ears hear
+  through), the keeper's `mode` (`input-watch`, or `polling system_profiler` and why),
   the one that serves speech on `127.0.0.1:8890`, and the home in use.
 
 ## Use
@@ -253,11 +260,17 @@ at `http://127.0.0.1:<port>/`, never at `localhost`. The code is `src/talk/`.
 `ears services install --only talk` keeps it running as a launch agent
 (opt-in; a bare install never does).
 
+In the kit's web page, Talk is a room: the kit's front door runs
+`talk serve --kit` on a unix socket and logs you in, so there is nothing to
+pair and you never run `--kit` by hand
+([docs/talk.md](docs/talk.md#in-the-kits-web-page)).
+
 ## Install
 
 The distribution is `sessionvoice` (`sessionvoice[nemotron]`, `sessionvoice[all]`);
-it installs the `ears`, `ears-serve`, `mouth` and `talk` commands. A bare
-`pip install sessionvoice` carries what `mouth say` needs. The extras are
+it installs the `ears`, `ears-serve`, `mouth`, `talk` and `sessionvoice-web`
+commands. A bare `pip install sessionvoice` carries what `mouth say` needs.
+The extras are
 `capture` (the mic and call ears), `silero` (the Silero VAD), `smart-turn`
 (the Smart Turn gate's ONNX backend), `nemotron` (mlx-audio) and `all`. An
 import that needs one fails naming it, as `pip install 'sessionvoice[EXTRA]'`.
@@ -295,7 +308,7 @@ with its fix, writes nothing and exits 2. A launch agent that fails to load
 also makes it exit 2, once every other service has been tried: one block on
 stderr names each service that did not load, the install log and `ears
 services status`. Every real run appends its output and launchctl's answers,
-timestamped, to `<sessionvoice's home>/logs/install.log`, and prints that path
+timestamped, to `install.log` in the log folder, `~/.sessionkit/voice/log/`, and prints that path
 as its `log` row ([logs](docs/logs.md#the-log-files)). It then:
 
 - first finds the mlx-audio speech server on `:8890`, or installs its own
@@ -346,7 +359,7 @@ as its `log` row ([logs](docs/logs.md#the-log-files)). It then:
   read (no `message`, or one older than 0.2.0), it warns, makes no box,
   checks neither, and goes on.
   On a postfix host it also prints the box to subscribe to:
-  `message subscription add heard+macbook`;
+  `message subscription add heard+room`;
 - `--agent NAME` adds NAME's wake seat (`ears wake add NAME`) and sends the
   keeper's notices to NAME (`EARS_NOTIFY`; `--notify ADDR` sends them
   elsewhere). Without it, and when the wake file names no seat, nobody would
@@ -411,7 +424,7 @@ the agent's next tool call. `EARS_HEARD_HOOK=off` silences it without
 editing the file.
 
 `ears heard hook` is the canonical heard hook. A seat that also gets heard
-mail (it subscribes to `heard+<device>`) gets each finished turn by mail, so
+mail (it subscribes to `heard+room`) gets each finished turn by mail, so
 give it `ears heard hook --partials-only || exit 1` instead: the hook then
 shows only the partial preview of a turn still being spoken and the shown
 events (wake word, barge-in, device change), never the finished turn a

{sessionvoice-0.3.1 → sessionvoice-0.3.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "sessionvoice"
-version = "0.3.1"
+version = "0.3.2"
 description = "One listening engine for coding agents and the Delta bot: Apple / Nemotron / whisper -> one event stream -> VAD -> turns"
 readme = "README.md"
 license = "AGPL-3.0-or-later"
@@ -44,6 +44,12 @@ ears = "ears.cli:main"
 ears-serve = "ears.serve:main"
 mouth = "mouth.__main__:main"
 talk = "talk.cli:main"
+# the kit's web front door asks it for Talk's room: `sessionvoice-web web-room`, src/talk/web.py
+sessionvoice-web = "talk.web:main"
+
+# the same room, for a front door installed in this environment: it loads and calls room()
+[project.entry-points."sessionkit.web"]
+talk = "talk.web:room"
 
 [dependency-groups]
 # what `make test` runs on: `uv sync` installs this group by default. The capture
@@ -64,8 +70,9 @@ where = ["src"]
 [tool.setuptools.package-data]
 # What the keeper agent runs, shipped so `ears services install` from a wheel points it at a
 # real script: the keeper, its call ears (a child) and the input watcher it
-# builds, and the turn mailer the mail agent runs (stdlib only, nothing beside it). The sounds
+# builds, and the turn mailer the mail agent runs (stdlib only). The sounds
 # agents' two watchers, and held_stream.py, which ears-sounds.py imports from beside itself.
+# phone_roots.py: where a Delta bot keeps its call sockets, imported by call-ears and ears-mail.py.
 # src/ears/_keeper/ holds symlinks into keeper/; the build copies the files.
 # The speaker side-car: run.sh, the side-car, the enrol's cluster-room.py and
 # make-refs.py, and the service venv's requirements.lock; src/ears/_speakers/ links into
@@ -76,7 +83,12 @@ where = ["src"]
 # Apple helper from a source checkout only, and refuses without one.
 ears = ["_keeper/ears-keeper.sh", "_keeper/call-ears", "_keeper/input-watch.swift", "_keeper/ears-mail.py",
         "_keeper/ears-sounds.py", "_keeper/mouth-sounds.py", "_keeper/held_stream.py",
+        "_keeper/phone_roots.py",
         "_speakers/*", "_speech/*", "_settings/*.schema.json"]
+# The separator's runner and its venv's lock: src/mouth/_separator/ links into separator/ the
+# same way. They are data, never imported by the mouth: `mouth split` runs run.py in the
+# separator venv. torch and demucs, that venv's, are in no dependency list here.
+mouth = ["_separator/*"]
 # talk serves its page and manifest from beside server.py
 talk = ["index.html", "manifest.json"]
 

{sessionvoice-0.3.1 → sessionvoice-0.3.2}/src/ears/_keeper/call-ears

@@ -23,6 +23,12 @@ as a child and restarts it, so the room and the calls are one job.
 import json, os, re, socket, subprocess, sys, time
 from pathlib import Path
 
+# phone_roots.py sits beside this file (keeper/, or ears/_keeper/ in a wheel); a loader
+# that runs this without its dir on sys.path still finds it.
+if os.path.dirname(os.path.abspath(__file__)) not in sys.path:
+    sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+import phone_roots  # noqa: E402  (keeper/ is not a package)
+
 HOME = Path.home()
 # The keeper passes its own; the default is the keeper's (keeper/ears-keeper.sh), change both.
 DIR = Path(os.environ.get("CALL_EARS_DIR")
@@ -58,11 +64,8 @@ def socks():
     env = os.environ.get("EARS_CALL_SOCKS", os.environ.get("VOICEMODE_CALL_EARS_SOCKS"))
     if env:
         return [p for p in env.split(":") if p]
-    sp = Path(os.environ.get("SESSIONPHONE_ROOT") or HOME / ".sessionphone").expanduser()
-    return sorted({str(p) for p in HOME.glob(".*/delta-chat/call-ears.sock")}
-                  | {str(p) for p in HOME.glob(".*/delta-chat/*/call-ears.sock")}
-                  | {str(p) for p in sp.glob("call-ears.sock")}
-                  | {str(p) for p in sp.glob("*/call-ears.sock")})
+    # else every bot root's: phone_roots.py, beside this file (the rule ears-mail.py shares)
+    return phone_roots.bot_socks("call-ears.sock")
 
 
 def live_calls():

{sessionvoice-0.3.1 → sessionvoice-0.3.2}/src/ears/_keeper/ears-keeper.sh

@@ -94,16 +94,23 @@ fi
 NOTIFY="${EARS_NOTIFY:-}"
 # Paths: settings in the voice settings folder, VOICE_SETTINGS (a copy of
 # ears.home.SETTINGS; tests/test_settings_dir.py holds it to the module),
-# logs in ~/Library/Logs/<thing>/: ~/Library/Logs/ears, as `ears services` (src/ears/
-# services.py LOGS) has it (a host may set EARS_LOG_DIR to keep
-# another log dir). RUN (stop files, the input-watch binary
+# logs in the voice log folder, asked of the ONE resolver (ears.home.log_dir, as `ears
+# services` (src/ears/services.py LOGS) has it; $EARS_LOG_DIR moves it, and the resolver
+# reads that too). No copy of the path here: a python that cannot answer stops the keeper
+# (exit 2) before it writes anything. RUN (stop files, the input-watch binary
 # and state) has no fixed home yet: it defaults beside the package's other assets
 # in ~/.cache/ears, and EARS_RUN moves it.
 VOICE_SETTINGS="$HOME/.sessionkit/voice/settings"
 CONF="${EARS_CONF:-$VOICE_SETTINGS/ears.json}"
-LOGDIR="${EARS_LOG_DIR:-$HOME/Library/Logs/ears}"
+LOGDIR="$("$PY" -m ears.home log-dir)" && [ -n "$LOGDIR" ] || { echo "ears-keeper: $PY -m ears.home log-dir named no log folder - run \`uv sync\` in $ROOT, or set EARS_PY" >&2; exit 2; }
 RUN="${EARS_RUN:-$HOME/.cache/ears/run}"
 mkdir -p "$LOGDIR" "$RUN"
+# The settings-schema copies (.ears.schema.json and the rest), brought up to date with the
+# schemas this python's package ships, once per start: an upgrade replaces the schemas, never
+# the copies. A copy whose bytes already match is left alone, and the log gets a line only
+# when one was written. It never stops the keeper.
+SCHEMA_LINE="$("$PY" -m ears.services schema --quiet 2>/dev/null || true)"
+[ -n "$SCHEMA_LINE" ] && echo "$(date +%H:%M:%S) $SCHEMA_LINE"
 SILERO="${EARS_SILERO_ONNX:-$HOME/.cache/ears/silero_vad.onnx}"   # where `ears install` puts it (src/ears/vad.py)
 WATCH_SRC="$(cd "$(dirname "$0")" && pwd)/input-watch.swift"
 WATCH_BIN="${EARS_WATCH_BIN:-$RUN/input-watch}"
@@ -506,6 +513,9 @@ deaf_subject() {  # $1 channel, $2 cause, $3 detail, $4 device -> the DEAF mail'
       if [ "$1" = room ]; then echo "$who DEAF, mic gives only zeros${4:+ on ${4:0:15}} at $(date +%H:%M) EOM"; return; fi ;;
     cover) # the side-car's: "ears: speakers DEAF, 0% of N lines on DEVICE EOM" (detail N, <= 3 digits)
       if [ "$1" = speakers ]; then echo "$who DEAF, 0% of ${3:0:3} lines${dev:+ on $dev} EOM"; return; fi ;;
+    speech) # the side-car's: "ears: speakers DEAF, no speech in N turns on DEVICE EOM" (detail N,
+      # <= 2 digits). The device is cut to 14 to keep 64.
+      if [ "$1" = speakers ]; then echo "$who DEAF, no speech in ${3:0:2} turns${4:+ on ${4:0:14}} EOM"; return; fi ;;
   esac
   max=$(( 64 - ${#head} - ${#tail} )); [ "$max" -ge 1 ] || max=1
   body="${body:0:$max}"; body="${body%% }"; body="${body%:}"   # no dangling ": " when cut
@@ -648,14 +658,21 @@ heard_events() {  # $1 = heard log, $2 = byte offset -> line 1 the offset after
   #   heard PID SESSION SOURCE DEVICE          a partial or turn with text: speech to text works.
   #                                            A run of the same line is printed once
   # With $3 = the side-car's speakers log and $4 = its byte offset, it reads that too, after them:
-  #   spk COVERED DEVICE                       a mic row (no `leg`: a call's rows are not the
-  #                                            room's). COVERED 0 = 0% of the line covered:
+  #   spk STATE DEVICE                         a mic row (no `leg`: a call's rows are not the
+  #                                            room's). STATE 0 = 0% of the line covered:
   #                                            `cover` == 0, or, before the side-car writes
   #                                            `cover`, a why ending "(0% of the line covered)".
-  #                                            1 = some: a row with no why was heard, and a
-  #                                            "no signal: the device gave only zeros" row says
-  #                                            "(N% ...)" with N > 0. A row that says neither
-  #                                            (a window in the future) is not printed
+  #                                            Some covered (a row with no why, or a why saying
+  #                                            "(N% ...)" with N > 0) is one of three:
+  #                                            1 = heard: no why, and a speaker named;
+  #                                            2 = covered and silent: a `turn` of 2 or more
+  #                                            words, no `mouth_utt`, whose why starts "words in
+  #                                            the line, no speech found" (the capture heard
+  #                                            words; the side-car's audio held no speech);
+  #                                            3 = covered, no evidence either way (zeros, a
+  #                                            partial, one word, the mouth's words, `silent`).
+  #                                            A row that says neither (a window in the future)
+  #                                            is not printed
   #   spkoff OFFSET                            last: the offset after its last whole line
   # Every field is non-empty ("?" for none): the TAB read would close up an empty one.
   python3 -c 'import json,re,sys
@@ -717,6 +734,10 @@ if len(sys.argv)>4:
     elif why.endswith("(0% of the line covered)"): z=0
     elif re.search(r"\(\d+% of the line covered\)$",why): z=1
     else: continue
+    if z==1 and (isinstance(why,str) or d.get("speaker") in ("silent","unheard")):
+      w="".join(ch.lower() if ch.isalnum() else " " for ch in str(d.get("text") or "")).split()
+      z=2 if (isinstance(why,str) and why.startswith("words in the line, no speech found")
+              and d.get("kind")=="turn" and len(w)>=2 and not d.get("mouth_utt")) else 3
     print("spk\t%d\t%s" % (z,txt(d.get("device"),20)))
   print("spkoff\t%d" % (so+se))' "$@"
 }
@@ -943,7 +964,7 @@ zeros_check() {  # the ONE reopen; call it every loop, after heard_watch (and be
   stop_capture || note "capture $zeros_pid did not stop"
 }
 # --- end zeros ---
-# --- speakers: the side-car hears nothing, 0% of its lines covered (tests/test_keeper_speakers_deaf.py runs this alone) ---
+# --- speakers: the side-car hears nothing: 0% of its lines covered, or covered with no speech (tests/test_keeper_speakers_deaf.py runs this alone) ---
 # The speaker side-car (speakers/speaker-sidecar2.py) opens the room capture's mic and names who
 # spoke each heard line. Its stream can open, stay active and deliver nothing: every line then gets
 # a row `why: "no audio from 'DEV' (0% of the line covered)"`. 27k such rows in 3h19m went unread
@@ -953,19 +974,37 @@ zeros_check() {  # the ONE reopen; call it every loop, after heard_watch (and be
 #   N lines on DEVICE`. The reopen is the side-car's own; the keeper only reports.
 # - A mic row with any coverage ends it: ONE recovery mail, `ears: speakers hearing again`. A
 #   "no signal: the device gave only zeros" row is covered (the room's zeros cause says that one).
+# - The other deaf: the stream is covered and its audio holds no speech while the capture heard
+#   words in the line. EARS_SPEAKERS_SILENT_TURNS (6) such turns in a row (state 2: a turn of 2+
+#   words, no mouth_utt; see heard_events) is DEAF, cause `speech`: ONE mail, `ears: speakers
+#   DEAF, no speech in N turns on DEVICE`. 6 because the side-car reopens its stream at 3 of
+#   them: the keeper reports what the reopen did not cure. A row with no words is no evidence,
+#   so a quiet room never counts. Only a heard row (state 1) ends this run and clears `speech`;
+#   a 0% row or a row that is no evidence (3) leaves it, so the two causes cannot hide each
+#   other. Every covered row (1, 2, 3) ends the 0% run and clears `cover`, as before.
 # - MUTED while the room is DEAF or no room capture runs: the side-car hears only while the keeper's
 #   capture runs, so its 0% rows are then a consequence, not a second fault. A muted row counts
-#   nothing and starts the run again. "No room capture" is the beat region's beat_pid: what
+#   nothing and starts its run again. "No room capture" is the beat region's beat_pid: what
 #   capture_pid said at the top of this loop (beat_track), so no row costs a second ps.
 SPK_DEAF_LINES="${EARS_SPEAKERS_DEAF_LINES:-20}"
 [[ "$SPK_DEAF_LINES" =~ ^[1-9][0-9]{0,2}$ ]] || SPK_DEAF_LINES=20   # 1-999: N fits the subject
-spk_n=0
+SPK_SILENT_TURNS="${EARS_SPEAKERS_SILENT_TURNS:-6}"
+[[ "$SPK_SILENT_TURNS" =~ ^[1-9][0-9]?$ ]] || SPK_SILENT_TURNS=6   # 1-99: N fits the subject
+spk_n=0; spk_q=0
 spk_muted() { deaf_room || [ -z "${beat_pid:-}" ]; }
-spk_seen() {  # $1 = 0 (a mic row with 0% of its line covered) or 1 (some covered), $2 = its device
-  if [ "$1" != 0 ]; then spk_n=0; deaf_clear speakers cover "$2"; return 0; fi
-  if spk_muted; then spk_n=0; return 0; fi
-  spk_n=$(( spk_n + 1 ))
-  [ "$spk_n" -ge "$SPK_DEAF_LINES" ] && deaf_set speakers cover "$SPK_DEAF_LINES" "$2"
+spk_seen() {  # $1 = the row's state (heard_events: 0 none covered, 1 heard, 2 covered and
+  # silent, 3 covered, no evidence), $2 = its device
+  case "$1" in
+    0) if spk_muted; then spk_n=0; return 0; fi
+       spk_n=$(( spk_n + 1 ))
+       [ "$spk_n" -ge "$SPK_DEAF_LINES" ] && deaf_set speakers cover "$SPK_DEAF_LINES" "$2" ;;
+    1) spk_n=0; spk_q=0; deaf_clear speakers cover "$2"; deaf_clear speakers speech "$2" ;;
+    2) spk_n=0; deaf_clear speakers cover "$2"
+       if spk_muted; then spk_q=0; return 0; fi
+       spk_q=$(( spk_q + 1 ))
+       [ "$spk_q" -ge "$SPK_SILENT_TURNS" ] && deaf_set speakers speech "$SPK_SILENT_TURNS" "$2" ;;
+    *) spk_n=0; deaf_clear speakers cover "$2" ;;
+  esac
   return 0
 }
 # --- end speakers ---