future/smart-speaker
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: improve streaming, alarms, and AI TTS

Browse Source
This commit is contained in:
future
2026-03-12 12:38:35 +03:00
parent 167ddc9264
commit 6769486e83
3 changed files with 171 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
app/core/ai.py
View File

@@ -283,11 +283,7 @@ def _extract_response_content(cfg, data: dict) -> str:
def _iter_openai_compatible_stream(response):
for line in response.iter_lines(decode_unicode=True):
if not line or not line.startswith("data:"):
continue
data_str = line[5:].strip()
for data_str in _iter_sse_data_lines(response):
if data_str == "[DONE]":
break
@@ -317,11 +313,7 @@ def _iter_openai_compatible_stream(response):
def _iter_anthropic_stream(response):
for line in response.iter_lines(decode_unicode=True):
if not line or not line.startswith("data:"):
continue
data_str = line[5:].strip()
for data_str in _iter_sse_data_lines(response):
if data_str == "[DONE]":
break
@@ -344,6 +336,28 @@ def _iter_anthropic_stream(response):
yield str(text)
def _iter_sse_data_lines(response):
"""
Читает SSE-стрим и возвращает только payload после "data:".
Явно декодируем как UTF-8, чтобы избежать mojibake вида "Пр...".
"""
for raw_line in response.iter_lines(decode_unicode=False):
if not raw_line:
continue
if isinstance(raw_line, bytes):
line = raw_line.decode("utf-8", errors="replace")
else:
line = str(raw_line)
if not line.startswith("data:"):
continue
data_str = line[5:].strip()
if data_str:
yield data_str
def _iter_stream_chunks(cfg, response):
if cfg["protocol"] == "anthropic":
yield from _iter_anthropic_stream(response)

150
app/features/alarm.py
View File

@@ -18,6 +18,121 @@ ALARM_FILE = BASE_DIR / "data" / "alarms.json"
ALARM_SOUND = BASE_DIR / "assets" / "sounds" / "Apex-1.mp3"
ASK_ALARM_TIME_PROMPT = "На какое время мне поставить будильник?"
_NUMBER_UNITS = {
"ноль": 0,
"один": 1,
"одна": 1,
"два": 2,
"две": 2,
"три": 3,
"четыре": 4,
"пять": 5,
"шесть": 6,
"семь": 7,
"восемь": 8,
"девять": 9,
}
_NUMBER_TEENS = {
"десять": 10,
"одиннадцать": 11,
"двенадцать": 12,
"тринадцать": 13,
"четырнадцать": 14,
"пятнадцать": 15,
"шестнадцать": 16,
"семнадцать": 17,
"восемнадцать": 18,
"девятнадцать": 19,
}
_NUMBER_TENS = {
"двадцать": 20,
"тридцать": 30,
"сорок": 40,
"пятьдесят": 50,
}
_PARTS_OF_DAY = {"утра", "дня", "вечера", "ночи"}
_FILLER_WORDS = {"мне", "меня", "пожалуйста", "на", "в", "во", "к", "и"}
_HOUR_WORDS = {"час", "часа", "часов"}
_MINUTE_WORDS = {"минута", "минуту", "минуты", "минут"}
def _parse_number_tokens(tokens, start_index: int):
if start_index >= len(tokens):
return None, 0
token = tokens[start_index]
if token.isdigit():
return int(token), 1
if token in _NUMBER_TEENS:
return _NUMBER_TEENS[token], 1
if token in _NUMBER_TENS:
value = _NUMBER_TENS[token]
if start_index + 1 < len(tokens):
next_token = tokens[start_index + 1]
if next_token in _NUMBER_UNITS:
value += _NUMBER_UNITS[next_token]
return value, 2
return value, 1
if token in _NUMBER_UNITS:
return _NUMBER_UNITS[token], 1
return None, 0
def _apply_part_of_day(hour: int, part_of_day: str | None) -> int:
if not part_of_day:
return hour
if part_of_day == "утра":
return 0 if hour == 12 else hour
if part_of_day == "ночи":
return 0 if hour == 12 else hour
if part_of_day in {"дня", "вечера"} and hour < 12:
return hour + 12
return hour
def _extract_alarm_time_words(text: str):
tokens = re.findall(r"[a-zа-я0-9]+", text.lower().replace("ё", "е"))
markers = {"будильник", "разбуди", "поставь", "установи", "включи", "на", "в", "к"}
for index, token in enumerate(tokens):
if token not in markers:
continue
current = index + 1
while current < len(tokens) and tokens[current] in _FILLER_WORDS:
current += 1
hour, consumed = _parse_number_tokens(tokens, current)
if hour is None:
continue
current += consumed
if current < len(tokens) and tokens[current] in _HOUR_WORDS:
current += 1
minute = 0
if current < len(tokens) and tokens[current] not in _PARTS_OF_DAY:
parsed_minute, minute_consumed = _parse_number_tokens(tokens, current)
if parsed_minute is not None:
minute = parsed_minute
current += minute_consumed
if current < len(tokens) and tokens[current] in _MINUTE_WORDS:
current += 1
part_of_day = None
if current < len(tokens) and tokens[current] in _PARTS_OF_DAY:
part_of_day = tokens[current]
if 0 <= hour <= 23 and 0 <= minute <= 59:
return _apply_part_of_day(hour, part_of_day), minute
return None
class AlarmClock:
def __init__(self):
@@ -70,10 +185,10 @@ class AlarmClock:
if re.search(r"\b(каждый день|ежедневно)\b", text):
return [0, 1, 2, 3, 4, 5, 6]
if re.search(r"\b(по будн|в будн|будние)\b", text):
if re.search(r"\b(?:по\s+будн\w*|в\s+будн\w*|будн\w*)\b", text):
days.update([0, 1, 2, 3, 4])
if re.search(r"\b(по выходн|в выходн|выходные)\b", text):
if re.search(r"\b(?:по\s+выходн\w*|в\s+выходн\w*|выходн\w*)\b", text):
days.update([5, 6])
day_patterns = {
@@ -268,32 +383,32 @@ class AlarmClock:
days = self._extract_alarm_days(text)
# Поиск формата "7:30", "7.30"
match = re.search(r"\b(\d{1,2})[:.-](\d{2})\b", text)
# Поиск формата "7:30", "7.30" и вариантов с "в/на/к".
match = re.search(r"(?:\b(?:на|в|во|к)\s+)?(\d{1,2})[:.-](\d{2})\b", text)
if match:
h, m = int(match.group(1)), int(match.group(2))
period_match = re.search(
r"\b(?:на|в|во|к)?\s*" + re.escape(match.group(0).strip()) + r"\s+(утра|дня|вечера|ночи)\b",
text,
)
part_of_day = period_match.group(1) if period_match else None
h = _apply_part_of_day(h, part_of_day)
if 0 <= h <= 23 and 0 <= m <= 59:
self.add_alarm_with_days(h, m, days=days)
days_phrase = self._format_days_phrase(days)
suffix = f" {days_phrase}" if days_phrase else ""
return f"Я установил будильник на {h} часов {m} минут{suffix}."
# Поиск формата словами "на 7 часов 15 минут"
# Поиск формата цифрами: "в 7 утра", "на 7", "к 6 30"
match_time = re.search(
r"на\s+(\d{1,2})(?:\s*(?:часов|часа|час))?(?:\s+(\d{1,2})(?:\s*(?:минут|минуты|минута))?)?",
r"(?:\b(?:на|в|во|к)\s+)?(\d{1,2})(?:\s*(?:часов|часа|час))?(?:\s+(\d{1,2})(?:\s*(?:минут|минуты|минута))?)?(?:\s+(утра|дня|вечера|ночи))?\b",
text,
)
if match_time:
h = int(match_time.group(1))
m = int(match_time.group(2)) if match_time.group(2) else 0
# Умная коррекция времени (если говорят "в 8", а сейчас 9, то это скорее 8 вечера или 8 утра завтра)
# Здесь простая логика AM/PM
if "вечера" in text and h < 12:
h += 12
elif "утра" in text and h == 12:
h = 0
h = _apply_part_of_day(h, match_time.group(3))
if 0 <= h <= 23 and 0 <= m <= 59:
self.add_alarm_with_days(h, m, days=days)
@@ -301,6 +416,15 @@ class AlarmClock:
suffix = f" {days_phrase}" if days_phrase else ""
return f"Хорошо, разбужу вас в {h}:{m:02d}{suffix}."
# Поиск формата словами: "в семь утра", "будильник семь тридцать"
word_time = _extract_alarm_time_words(text)
if word_time:
h, m = word_time
self.add_alarm_with_days(h, m, days=days)
days_phrase = self._format_days_phrase(days)
suffix = f" {days_phrase}" if days_phrase else ""
return f"Хорошо, разбужу вас в {h}:{m:02d}{suffix}."
if re.search(r"(постав|установ|запусти|включи|разбуди)", text) or text.strip() in {
"будильник",
"поставь будильник",

68
app/main.py
View File

@@ -2,14 +2,9 @@
Smart Speaker - Main Application
"""
import os
import os
import queue
import re
import signal
import sys
import threading
import time
from collections import deque
@@ -566,38 +561,8 @@ def main():
# AI запрос
chat_history.append({"role": "user", "content": user_text})
# Очередь для TTS
tts_q = queue.Queue()
interrupt_event = threading.Event()
def tts_worker():
"""Фоновый поток для озвучки."""
while True:
item = tts_q.get()
if item is None:
tts_q.task_done()
break
text, lang = item
if interrupt_event.is_set():
tts_q.task_done()
continue
completed = speak(
text, check_interrupt=check_wakeword_once, language=lang
)
if not completed:
interrupt_event.set()
tts_q.task_done()
worker_thread = threading.Thread(target=tts_worker, daemon=True)
worker_thread.start()
full_response = ""
buffer = ""
interrupted = False
try:
# Streaming от AI
@@ -606,33 +571,20 @@ def main():
print("🤖 AI: ", end="", flush=True)
for chunk in stream_generator:
if interrupt_event.is_set():
break
buffer += chunk
full_response += chunk
print(chunk, end="", flush=True)
# Конец предложения
if re.search(r"[.!?\n]+(?:\s|$)", buffer):
clean_chunk = clean_response(buffer, language="ru")
if clean_chunk.strip():
tts_q.put((clean_chunk, "ru"))
buffer = ""
# Остаток
if buffer.strip() and not interrupt_event.is_set():
clean_chunk = clean_response(buffer, language="ru")
if clean_chunk.strip():
tts_q.put((clean_chunk, "ru"))
except Exception as e:
print(f"\n❌ Ошибка: {e}")
speak("Произошла ошибка при получении ответа.")
# Ждем окончания озвучки
tts_q.put(None)
worker_thread.join()
else:
clean_ai_response = clean_response(full_response, language="ru")
if clean_ai_response.strip():
interrupted = not speak(
clean_ai_response,
check_interrupt=check_wakeword_once,
language="ru",
)
print()
@@ -643,7 +595,7 @@ def main():
stop_wakeword_monitoring()
skip_wakeword = True
if interrupt_event.is_set():
if interrupted:
print("⏹️ Ответ прерван")
print()