whoops

.gitignore

@@ -1,5 +1,7 @@
 **/*.sw*
 /whisper-cpp-2023/rust.d/target
+/rust-whisper.d/target
+/rust-whisper.d/models
 snowboy-2022/snowboy
 **/*.git.d
 **/*.wav

rust-whisper.d/download_models.sh

@@ -0,0 +1,10 @@
+#!/bin/bash
+
+src="https://huggingface.co/ggerganov/whisper.cpp/resolve/main/ggml"
+d="${1:-"$PWD"/models}"
+mkdir -p "$d"
+
+# Whisper models
+for model in "tiny.en" "base.en" "small.en" "medium.en"; do
+    	test -f "$d"/ggml-$model.bin || wget --quiet --show-progress -O "$d"/ggml-$model.bin "$src-$model.bin"
+done

rust-whisper.d/hotwords.py

@@ -52,32 +52,55 @@ class Reader(threading.Thread):
       self.inq = inq
       self.outq = outq
 
-   def run(self):
-      log("Reader.run: start")
-      try:
-         idx = [
+   def mic_idx(self):
+      mics = []
+      while not mics and not self.should_stop():
+         log(f'searching for one of {self.name.split(",")} in {sr.Microphone.list_microphone_names()}...')
+         time.sleep(1)
+         mics = [
             idx for idx,v in enumerate(
                sr.Microphone.list_microphone_names(),
             ) if v in self.name.split(",")
-         ][0]
-         with sr.Microphone(device_index=idx) as mic:
-            while not self.should_stop():
-               try:
-                  self.outq.put(self._run(mic))
-               except Exception as e:
-                  if not "timed out" in str(e):
-                     log("Reader.run: error:", e)
+         ]
+      log("mic#", mics[0])
+      return mics[0]
+
+   def run(self):
+      try:
+         log("Reader.run: start")
+         self._run()
       except Exception as e:
-         log("Reader.run panic:", e)
-         log("microphones:", sr.Microphone.list_microphone_names())
+         log("Reader:run: exit:", e)
       finally:
          self.outq.put(None)
-      log("Reader.run: stop")
+         log("Reader.run: stop")
+
+   def _run(self):
+      while not self.should_stop():
+         time.sleep(3)
+         mic = sr.Microphone(device_index=self.mic_idx())
+         try:
+            mic.__enter__()
+            while not self.should_stop():
+               try:
+                  self.outq.put(self.read(mic))
+               except Exception as e:
+                  if not "timed out" in str(e):
+                     raise e
+         except Exception as e:
+            import traceback
+            traceback.print_exception(e)
+            log("Reader.run: error:", e)
+         finally:
+            try:
+               mic.__exit__(None, None, None)
+            except Exception as e:
+               log("Reader.run.catch: error:", e)
 
    def should_stop(self):
       return not self.inq.empty()
 
-   def _run(self, mic):
+   def read(self, mic):
       mic_timeout = int(os.environ.get("MIC_TIMEOUT", 5))
       r = sr.Recognizer()
       return r.listen(
@@ -116,6 +139,7 @@ class Parser(threading.Thread):
       proc = subprocess.run(f"MODEL=./models/ggml-{os.environ.get('MODEL','tiny.en')}.bin WAV={p} P=2 rust-whisper", capture_output=True, shell=True)
       result = proc.stdout.decode().strip()
       if os.environ.get("DEBUG", None):
+         log("stderr:", proc.stderr.decode().strip())
          log("raw transcript:", result)
       result = result.replace(">>", "")
       result = "".join([i.split("]")[-1] for i in result.split("[")[0]])
@@ -256,7 +280,7 @@ class Actor(threading.Thread):
          self.pid = int(environ["SIGUSR2"])
          self.handle = self.handle_signal
       elif os.environ.get("URL", ""):
-         self.url = environ["URL"]
+         self.url = os.environ["URL"]
          self.handle = self.handle_url
          self.headers = [i.split("=")[:2] for i in os.environ.get("HEADERS", "").split("//") if i]
          self.body = os.environ.get("BODY", '{"hotword":"{{hotword}}","context":"{{context}}"}')

rust-whisper.d/requirements.txt

@@ -2,3 +2,4 @@ git+https://github.com/openai/whisper.git
 soundfile
 PyAudio
 SpeechRecognition
+PyYAML

whisper-2023/hotwords.py

@@ -1,304 +0,0 @@
-import speech_recognition as sr
-import time
-import threading
-import queue
-import signal
-import sys
-import os
-import requests
-import yaml
-
-def log(*args):
-   print(">", *args, file=sys.stderr)
-
-class Piper(threading.Thread):
-   def __init__(self, inq, outq):
-      threading.Thread.__init__(self)
-      self.inq = inq
-      self.outq = outq
-
-   def run(self):
-      while True:
-         got = self.inq.get()
-         if got is None:
-            break
-         self._run(got)
-      self.outq.put(None)
-
-class Manager(threading.Thread):
-   def __init__(self, outq):
-      threading.Thread.__init__(self)
-      self.outq = outq
-      inq = queue.Queue()
-      def catcher(sig, frame):
-         inq.put(None)
-      self.inq = inq
-      signal.signal(signal.SIGINT, catcher)
-
-   def run(self):
-      log("Manager.run: start")
-      self.inq.get()
-      self.outq.put(None)
-      log("Manager.run: stop")
-
-class Reader(threading.Thread):
-   def __init__(self, inq, outq):
-      threading.Thread.__init__(self)
-      self.name = os.environ.get("MIC_NAME", "pulse_monitor")
-      if not self.name:
-         for index, name in enumerate(sr.Microphone.list_microphone_names()):
-             print("[{0}] Microphone with name \"{1}\" found for `Microphone(device_index={0})`".format(index, name))
-         exit()
-      self.inq = inq
-      self.outq = outq
-
-   def run(self):
-      log("Reader.run: start")
-      try:
-         idx = [
-            idx for idx,v in enumerate(
-               sr.Microphone.list_microphone_names(),
-            ) if v in self.name.split(",")
-         ][0]
-         with sr.Microphone(device_index=idx) as mic:
-            while not self.should_stop():
-               try:
-                  self.outq.put(self._run(mic))
-               except Exception as e:
-                  if not "timed out" in str(e):
-                     log("Reader.run: error:", e)
-      except Exception as e:
-         log("Reader.run panic:", e)
-         log("microphones:", sr.Microphone.list_microphone_names())
-      finally:
-         self.outq.put(None)
-      log("Reader.run: stop")
-
-   def should_stop(self):
-      return not self.inq.empty()
-
-   def _run(self, mic):
-      mic_timeout = int(os.environ.get("MIC_TIMEOUT", 5))
-      r = sr.Recognizer()
-      return r.listen(
-         mic,
-         timeout=mic_timeout,
-         phrase_time_limit=mic_timeout,
-      )
-
-class Parser(threading.Thread):
-   def __init__(self, inq, outq):
-      threading.Thread.__init__(self)
-      self.inq = inq
-      self.outq = outq
-
-   def run(self):
-      log("Parser.run: start")
-      while True:
-         try:
-            clip = self.inq.get()
-            backlog = self.inq.qsize()
-            if backlog:
-               log("Parser.run backlog", backlog)
-            if clip is None:
-               break
-            self.outq.put(self._run(clip).strip())
-         except Exception as e:
-            log("Parser.run: error:", e)
-      self.outq.put(None)
-      log("Parser.run: stop")
-
-   def _run(self, clip):
-      r = sr.Recognizer()
-      return r.recognize_whisper(clip, language="english", model=os.environ.get("MODEL", "small.en")) # tiny.en=32x, base.en=16x, small.en=6x, medium.en=x2
-
-def load_dot_notation(v, s):
-   items = s.replace("[]", ".[]").split(".")
-   return _load_dot_notation(v, items)
-
-def _load_dot_notation(v, items):
-   for i in range(len(items)):
-      k = items[i]
-      if not k:
-         continue
-      if k == "[]":
-         if isinstance(v, list):
-            result = []
-            for j in v:
-               subresult = _load_dot_notation(j, items[i+1:])
-               if isinstance(subresult, list):
-                  result.extend(subresult)
-               else:
-                  result.append(subresult)
-            return result
-         else:
-            result = []
-            for j in v.values():
-               subresult = _load_dot_notation(j, items[i+1:])
-               if isinstance(subresult, list):
-                  result.extend(subresult)
-               else:
-                  result.append(subresult)
-            return result
-      else:
-         if isinstance(v, list):
-            v = v[int(k)]
-         else:
-            v = v[k]
-   return v
-
-def test_load_dot_notation():
-   for i in [
-      "a" == load_dot_notation("a", "."),
-      ["a"] == load_dot_notation(["a"], "."),
-      "b" == load_dot_notation({"a":"b"}, ".a"),
-      "c" == load_dot_notation({"a":{"b":"c"}}, ".a.b"),
-      "c" == load_dot_notation({"a":{"b":["c"]}}, ".a.b.0"),
-      ["c","d"] == load_dot_notation({"a":{"b":"c"}, "a2":{"b":"d"}}, ".[].b"),
-      ["c","d"] == load_dot_notation({"a":{"b":["c"], "b2":["d"]}}, ".a.[].0"),
-      ["c","d"] == load_dot_notation({"a":{"b":["c"], "b2":["d"]}}, ".a[].0"),
-      ["c","d"] == load_dot_notation(["c", "d"], "."),
-      ["c","d"] == load_dot_notation(["c", "d"], "[]"),
-   ]:
-      if not i:
-         raise Exception(i)
-test_load_dot_notation()
-
-class Reactor(threading.Thread):
-   def __init__(self, inq, outq):
-      threading.Thread.__init__(self)
-      self.inq = inq
-      self.outq = outq
-      self.load_hotwords = Reactor.new_load_hotwords()
-      log(f"hotwords: {self.load_hotwords()}")
-
-   def new_load_hotwords():
-      p = os.environ.get("HOTWORDS", None)
-      if not p:
-         def load_nothing():
-            return []
-         return load_nothing
-
-      try:
-         if "@" in p:
-            def load_hotwords_in_yaml_file():
-               with open(p.split("@")[0], "r") as f:
-                  v = yaml.safe_load(f)
-                  v = load_dot_notation(v, p.split("@")[-1])
-                  return ["".join(i.strip().lower().split()) for i in v if i]
-            load_hotwords_in_yaml_file()
-            return load_hotwords_in_yaml_file
-         else:
-            def load_hotwords_in_file():
-               with open(p, "r") as f:
-                  return ["".join(i.strip().lower().split()) for i in f.readlines()]
-            load_hotwords_in_file()
-            return load_hotwords_in_file
-      except Exception as e:
-         log(f"$HOTWORDS {p} is not a file: {e}")
-
-      hotwords = ["".join(i.lower().strip().split()) for i in p.split("\/\/")]
-      log(f'$HOTWORDS: {hotwords}')
-      def load_hotwords_as_literal():
-         return hotwords
-      return load_hotwords_as_literal
-
-   def run(self):
-      log("Reactor.run: start")
-      while True:
-         text = self.inq.get()
-         if text is None:
-            break
-         self.handle(text)
-      self.outq.put(None)
-      log("Reactor.run: stop")
-
-   def handle(self, text):
-      hotwords = self.load_hotwords()
-      if os.environ.get("DEBUG", None):
-         log(f"seeking {hotwords} in {text}")
-      if not hotwords:
-         if not os.environ.get("HOTWORDS", None):
-            print(text)
-         else:
-            log(text)
-         return
-      cleantext = "".join([i for i in "".join(text.lower().split()) if i.isalpha()])
-      for i in hotwords:
-         if i in cleantext:
-            #log(f"Reactor.handle: found hotword '{i}' in '{text}' as '{cleantext}'")
-            self.outq.put((i, text))
-
-class Actor(threading.Thread):
-   def __init__(self, inq):
-      threading.Thread.__init__(self)
-      self.inq = inq
-      self.handle = self.handle_stderr
-      if os.environ.get("STDOUT", "") == "true":
-         self.handle = self.handle_stdout
-      elif os.environ.get("SIGUSR2", ""):
-         self.pid = int(environ["SIGUSR2"])
-         self.handle = self.handle_signal
-      elif os.environ.get("URL", ""):
-         self.url = environ["URL"]
-         self.handle = self.handle_url
-         self.headers = [i.split("=")[:2] for i in os.environ.get("HEADERS", "").split("//") if i]
-         self.body = os.environ.get("BODY", '{"hotword":"{{hotword}}","context":"{{context}}"}')
-         log(self.headers)
-
-   def run(self):
-      log("Actor.run: start")
-      while True:
-         got = self.inq.get()
-         if got is None:
-            break
-         self.handle(got[0], got[1])
-      log("Actor.run: stop")
-
-   def handle_stderr(self, hotword, context):
-      log(f"'{hotword}' in '{context}'")
-
-   def handle_stdout(self, hotword, context):
-      log(context)
-      print(hotword)
-
-   def handle_signal(self, hotword, context):
-      self.handle_stderr(hotword, context)
-      os.kill(self.pid, signal.SIGUSR2)
-
-   def handle_url(self, hotword, context):
-      self.handle_stderr(hotword, context)
-      try:
-         headers = {}
-         for i in self.headers:
-            key = i[0]
-            value = i[1]
-            value = value.replace("{{hotword}}", hotword)
-            value = value.replace("{{context}}", context)
-            headers[key] = value
-         body = self.body
-         body = body.replace("{{hotword}}", hotword)
-         body = body.replace("{{context}}", context)
-         if os.environ.get("DEBUG", "") :
-            log("POST", self.url, headers, body)
-         requests.post(self.url, headers=headers, data=body)
-      except Exception as e:
-         log("Actor.handle_url:", e)
-
-def main():
-   managerToParserQ = queue.Queue(maxsize=1)
-   readerToParserQ = queue.Queue(maxsize=10)
-   parserToReactorQ = queue.Queue(maxsize=10)
-   reactorToActorQ = queue.Queue(maxsize=10)
-   threads = [
-      Manager(managerToParserQ),
-      Reader(managerToParserQ, readerToParserQ),
-      Parser(readerToParserQ, parserToReactorQ),
-      Reactor(parserToReactorQ, reactorToActorQ),
-      Actor(reactorToActorQ),
-   ]
-   [t.start() for t in threads]
-   [t.join() for t in threads]
-
-if __name__ == "__main__":
-   main()

whisper-2023/install.sh

@@ -1,7 +0,0 @@
-#! /bin/bash
-
-sudo apt install portaudio19-dev python3-pyaudio
-python3 -m pip install git+https://github.com/openai/whisper.git soundfile PyAudio SpeechRecognition
-
-#sudo apt-get install python3 python3-all-dev python3-pip build-essential swig git libpulse-dev libasound2-dev
-#python3 -m pip install pocketsphinx

whisper-cpp-2023/Dockerfile

@@ -1,20 +0,0 @@
-FROM debian:buster as builder
-
-RUN apt -y update && apt -y install build-essential wget ffmpeg
-
-WORKDIR /tmp/whisper-cpp.git.d
-RUN wget https://github.com/ggerganov/whisper.cpp/archive/refs/tags/v1.2.1.tar.gz \
-   && tar -xf ./*.tar.gz \
-   && mv ./whisper*/ ./git.d
-WORKDIR /tmp/whisper-cpp.git.d/git.d
-RUN make && make samples
-
-FROM debian:buster
-RUN apt -y update && apt -y install curl
-COPY --from=builder /tmp/whisper-cpp.git.d/git.d/ /whisper-cpp.git.d/
-WORKDIR /whisper.d
-RUN bash /whisper-cpp.git.d/models/download-ggml-model.sh tiny.en
-
-ENTRYPOINT []
-CMD /whisper-cpp.git.d/main -m /whisper-cpp.git.d/models/ggml-tiny.en.bin -f /whisper-cpp.git.d/samples/gb1.wav -t 4
-

whisper-cpp-2023/c-whisper

@@ -1 +0,0 @@
-git.d/main

whisper-cpp-2023/compare.md

@@ -1,14 +0,0 @@
-# git.d/samples/mm0.wav (30s)
-
-| model | threads | rust          | c           |
-| ----- | ------- | ------------- | ----------- |
-| tiny  | 1       | 4.4s@122%     | 4.9s@125%   |
-| tiny  | 2       | 2.7s@210%     | 3.3s@190%   |
-| tiny  | 4       | 2.0s@-        | 2.9s@400%   |
-| tiny  | 8       | 2.0s@-        | 3.1s@700%   |
-| small | 1       | 23.9s@175%    | 28.5s@205%  |
-| small | 2       | 14.9s@347%    | 19.2s@330%  |
-| small | 4       | 12.3s@515%    | 22.1s@530%  |
-| base  | 1       | 8.7s@150%     | 10.2s@155%  |
-| base  | 2       | 5.1s@240%     | 7.1s@270%   |
-| base  | 4       | 3.8s@370%     | 6.0s@430%   |

whisper-cpp-2023/compare.sh

@@ -1,7 +0,0 @@
-export P=${1:-1}
-export MODEL=${2:-models/ggml-tiny.en.bin}
-export WAV=${3:-git.d/samples/jfk.wav}
-echo === RUST
-time rust-whisper 2>&1 | grep -v ^whisper_ | grep ..
-echo === C
-time ./c-whisper -m $MODEL -f $WAV -t $P 2>&1 | grep -v ^whisper_ | grep -v ^system_info | grep -v ^main: | grep ..

whisper-cpp-2023/libwhisper.a

@@ -1 +0,0 @@
-git.d/libwhisper.a

whisper-cpp-2023/models

@@ -1 +0,0 @@
-git.d/models

whisper-cpp-2023/native.sh

@@ -1,24 +0,0 @@
-#! /bin/bash
-
-if [ ! -d ./git.d/.git ]; then
-   git clone https://github.com/ggerganov/whisper.cpp.git git.d
-fi
-
-cd ./git.d
-if [ ! -f ./samples/gb1.wav ]; then
-   make samples
-fi
-if [ ! -f ./main ]; then
-   make
-fi
-if [ ! -f ./stream ]; then
-   make stream
-fi
-if [ ! -f ./models/ggml-${MODEL:-tiny.en}.bin ]; then
-   bash ./models/download-ggml-model.sh ${MODEL:-tiny.en}
-fi
-if [ -n "$STREAM" ]; then
-   ./stream -m ./models/ggml-${MODEL:-tiny.en}.bin -t 8 --step 500 --length ${MIC_TIMEOUT:-2}000 $(test -n "$MIC_ID" && echo -c "$MIC_ID")
-else
-   time ./main -m ./models/ggml-${MODEL:-tiny.en}.bin -f ./samples/gb1.wav -t 4
-fi

whisper-cpp-2023/whisper.h

@@ -1 +0,0 @@
-git.d/whisper.h