From 0efda5920470fb539950831ced7fbd799c3666c0 Mon Sep 17 00:00:00 2001
From: Joshua Yuen <joshuayuen99@gmail.com>
Date: Sat, 12 Sep 2026 12:46:30 -0400
Subject: [PATCH] Add hold-to-talk voice dictation

---
 modules/home/default.nix      |   1 +
 modules/home/dictation.nix    | 128 ++++++++++++++++++++++++++++++++++
 modules/options/default.nix   |   1 +
 modules/options/dictation.nix |  24 +++++++
 profiles/desktop/default.nix  |   2 +
 5 files changed, 156 insertions(+)
 create mode 100644 modules/home/dictation.nix
 create mode 100644 modules/options/dictation.nix

diff --git a/modules/home/default.nix b/modules/home/default.nix
index 1c84110..9e9faec 100644
--- a/modules/home/default.nix
+++ b/modules/home/default.nix
@@ -5,6 +5,7 @@
   imports = [
     ./direnv.nix
     ./discord.nix
+    ./dictation.nix
     ./firefox.nix
     ./gpg.nix
     ./fzf.nix
diff --git a/modules/home/dictation.nix b/modules/home/dictation.nix
new file mode 100644
index 0000000..294a68c
--- /dev/null
+++ b/modules/home/dictation.nix
@@ -0,0 +1,128 @@
+{
+  config,
+  lib,
+  pkgs,
+  ...
+}:
+with lib;
+let
+  cfg = config.modules.dictation;
+  hyprlandCfg = config.modules.hyprland;
+
+  models = {
+    "tiny.en" = pkgs.fetchurl {
+      url = "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-tiny.en.bin";
+      hash = "sha256-kh5M+Ghv3Zk9zQgaXaW2w2W/3hFi5ysI11rHUomSCx8=";
+    };
+    "base.en" = pkgs.fetchurl {
+      url = "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-base.en.bin";
+      hash = "sha256-oDd5yG3zMjB19eeWyyzlAp8A7Ihp7uP9+4l6/jbG0AI=";
+    };
+    "small.en" = pkgs.fetchurl {
+      url = "https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml-small.en.bin";
+      hash = "sha256-xhONbVjsyDIgl+D5h8MvG+i7ChhTKj+I9zTRu/nEHl0=";
+    };
+  };
+
+  runtimeDeps = with pkgs; [
+    coreutils
+    gnused
+    libnotify
+    pipewire
+    whisper-cpp
+    wtype
+  ];
+
+  dictation = pkgs.writeShellScriptBin "dictation" ''
+    PATH=${lib.makeBinPath runtimeDeps}:''${PATH:-}
+    set -euo pipefail
+
+    model="${models.${cfg.model}}"
+
+    dir="''${XDG_RUNTIME_DIR:-/tmp}/dictation"
+    mkdir -p "$dir"
+    pidfile="$dir/pid"
+    wav="$dir/clip.wav"
+
+    notify() {
+      notify-send -t 2500 "Dictation" "$1" 2>/dev/null || true
+    }
+
+    case "''${1:-}" in
+      start)
+        if [ -f "$pidfile" ] && kill -0 "$(cat "$pidfile")" 2>/dev/null; then
+          exit 0
+        fi
+        rm -f "$pidfile" "$wav"
+        notify "Listening..."
+        pw-record --rate 16000 --channels 1 --format s16 "$wav" >/dev/null 2>&1 &
+        pid=$!
+        echo "$pid" > "$pidfile"
+        disown
+        sleep 0.3
+        if ! kill -0 "$pid" 2>/dev/null; then
+          rm -f "$pidfile"
+          notify "Microphone unavailable"
+        fi
+        ;;
+      stop)
+        [ -f "$pidfile" ] || exit 0
+        pid="$(cat "$pidfile")"
+        rm -f "$pidfile"
+        if kill -0 "$pid" 2>/dev/null; then
+          kill -INT "$pid" 2>/dev/null || true
+          for _ in $(seq 1 40); do
+            kill -0 "$pid" 2>/dev/null || break
+            sleep 0.05
+          done
+          kill -0 "$pid" 2>/dev/null && kill -KILL "$pid" 2>/dev/null || true
+        fi
+        sleep 0.1
+
+        if [ ! -f "$wav" ]; then
+          notify "Recording missing"
+          exit 0
+        fi
+        # Ignore clips shorter than 0.5s (16 kHz, s16, mono: 32000 B/s + 44 B header)
+        size="$(stat -c %s "$wav" 2>/dev/null || echo 0)"
+        if [ "$size" -lt 16044 ]; then
+          rm -f "$wav"
+          exit 0
+        fi
+
+        if ! text="$(whisper-cli -np -nt -t "$(nproc)" -m "$model" -f "$wav" 2>/dev/null \
+          | sed -e 's/([^)]*)//g' -e 's/\[[^]]*\]//g' \
+          | tr '\n' ' ' \
+          | sed -e 's/  */ /g' -e 's/^ *//' -e 's/ *$//')"; then
+          rm -f "$wav"
+          notify "Transcription failed"
+          exit 1
+        fi
+        rm -f "$wav"
+
+        if [ -z "$text" ]; then
+          notify "No speech detected"
+          exit 0
+        fi
+
+        wtype -- "$text" || notify "Could not type transcription"
+        ;;
+      *)
+        echo "usage: dictation {start|stop}" >&2
+        exit 1
+        ;;
+    esac
+  '';
+in
+{
+  config = mkIf cfg.enable {
+    home.packages = [ dictation ];
+
+    wayland.windowManager.hyprland = mkIf hyprlandCfg.enable {
+      settings = {
+        bind = [ "SUPER, T, exec, dictation start" ];
+        bindr = [ "SUPER, T, exec, dictation stop" ];
+      };
+    };
+  };
+}
diff --git a/modules/options/default.nix b/modules/options/default.nix
index 34f71db..b7e1f7b 100644
--- a/modules/options/default.nix
+++ b/modules/options/default.nix
@@ -4,6 +4,7 @@
 {
   imports = [
     ./discord.nix
+    ./dictation.nix
     ./display.nix
     ./gnomeBoxes.nix
     ./gpg.nix
diff --git a/modules/options/dictation.nix b/modules/options/dictation.nix
new file mode 100644
index 0000000..fff43aa
--- /dev/null
+++ b/modules/options/dictation.nix
@@ -0,0 +1,24 @@
+{
+  lib,
+  ...
+}:
+with lib;
+{
+  options.modules.dictation = {
+    enable = mkEnableOption {
+      default = false;
+      description = "Enable hold-to-talk voice dictation (hold SUPER+T)";
+    };
+    model = mkOption {
+      type = types.enum [
+        "tiny.en"
+        "base.en"
+        "small.en"
+      ];
+      default = "base.en";
+      description = ''
+        The local Whisper model to use for transcription.
+      '';
+    };
+  };
+}
diff --git a/profiles/desktop/default.nix b/profiles/desktop/default.nix
index fbdaad2..74d72a2 100644
--- a/profiles/desktop/default.nix
+++ b/profiles/desktop/default.nix
@@ -22,5 +22,7 @@
 
   modules.discord.enable = true;
 
+  modules.dictation.enable = true;
+
   modules.tailscale.enable = true;
 }