Files
Bolt-Pet/tests/test_tts_stream.py
T

176 lines
6.6 KiB
Python

"""Streaming-TTS chunk reassembly and the near-miss log. Both are pure —
no network, no audio device, no ONNX model."""
import sys
from pathlib import Path
from unittest.mock import MagicMock, patch
import numpy as np
import pytest
sys.path.insert(0, str(Path(__file__).resolve().parent.parent))
from bolt_pet import config as tts_config
from bolt_pet.audio import tts
from bolt_pet.audio.tts import TtsError, chunks_to_int16, voice_for
from bolt_pet.audio.wake_word import NearMissLog
def _pcm(*values):
return np.array(values, dtype=np.int16).tobytes()
def test_whole_samples_pass_straight_through():
chunks = list(chunks_to_int16([_pcm(1, 2), _pcm(3, 4)]))
assert np.concatenate(chunks).tolist() == [1, 2, 3, 4]
def test_a_sample_split_across_two_http_chunks_is_rejoined():
# The killer bug this exists to prevent: an odd byte at a chunk boundary
# shifts everything after it by one byte and plays as static.
raw = _pcm(100, -200, 300, -400)
chunks = list(chunks_to_int16([raw[:3], raw[3:]]))
assert np.concatenate(chunks).tolist() == [100, -200, 300, -400]
def test_many_odd_boundaries_in_a_row():
raw = _pcm(*range(1, 21))
pieces = [raw[i:i + 3] for i in range(0, len(raw), 3)] # every boundary odd
assert np.concatenate(list(chunks_to_int16(pieces))).tolist() == list(range(1, 21))
def test_empty_chunks_are_skipped():
assert list(chunks_to_int16([b"", b""])) == []
def test_a_dangling_byte_at_the_end_is_dropped_not_played():
raw = _pcm(7, 8) + b"\x01"
assert np.concatenate(list(chunks_to_int16([raw]))).tolist() == [7, 8]
# ── wake-word near misses ───────────────────────────────────────────────────
def test_scores_just_under_the_threshold_are_recorded():
log = NearMissLog(limit=10, margin=0.2)
assert log.observe(0.45, threshold=0.5, timestamp=1.0) is True
assert log.entries() == [(1.0, 0.45, 0.5)]
def test_detections_and_background_noise_are_not_near_misses():
log = NearMissLog(limit=10, margin=0.2)
assert log.observe(0.90, threshold=0.5, timestamp=1.0) is False # it fired
assert log.observe(0.05, threshold=0.5, timestamp=2.0) is False # just noise
assert log.entries() == []
def test_peak_tracks_every_score_not_just_near_misses():
log = NearMissLog(limit=10, margin=0.2)
log.observe(0.30, threshold=0.5, timestamp=1.0)
log.observe(0.95, threshold=0.5, timestamp=2.0)
log.observe(0.10, threshold=0.5, timestamp=3.0)
assert log.peak == 0.95
def test_the_log_is_bounded():
log = NearMissLog(limit=3, margin=0.2)
for i in range(10):
log.observe(0.45, threshold=0.5, timestamp=float(i))
assert len(log.entries()) == 3
assert log.entries()[-1][0] == 9.0
def test_clear_resets_peak_and_entries():
log = NearMissLog(limit=3, margin=0.2)
log.observe(0.45, threshold=0.5, timestamp=1.0)
log.clear()
assert log.entries() == [] and log.peak == 0.0
# ── voice selection (server speak_as) ───────────────────────────────────────
def test_the_override_voice_wins_over_the_configured_one(monkeypatch):
monkeypatch.setattr(tts_config, "ELEVENLABS_VOICE_ID", "DEFAULT")
assert voice_for("VOICE1") == "VOICE1"
assert voice_for("") == "DEFAULT"
assert voice_for(None) == "DEFAULT"
# ── server-hosted synthesis (/desk/tts) ─────────────────────────────────────
# No local ElevenLabs account: both the whole-clip and streaming paths post
# to this pet's own server, same as server_client.py's other endpoints.
@pytest.fixture(autouse=True)
def _configured(monkeypatch):
monkeypatch.setattr(tts_config, "SERVER_URL", "http://test-server:5002")
monkeypatch.setattr(tts_config, "API_KEY", "test-key")
monkeypatch.setattr(tts_config, "SESSION_ID", "pet-test")
monkeypatch.setattr(tts_config, "ELEVENLABS_VOICE_ID", "default-voice")
def _mock_response(content=b"", ok=True):
resp = MagicMock()
resp.content = content
resp.raise_for_status = MagicMock() if ok else MagicMock(side_effect=Exception("boom"))
return resp
def test_synthesize_pcm_posts_to_the_servers_tts_endpoint():
pcm_bytes = np.array([1, 2, 3], dtype=np.int16).tobytes()
with patch.object(tts.requests, "post") as post:
post.return_value = _mock_response(pcm_bytes)
pcm, rate = tts.synthesize_pcm("hello there")
post.assert_called_once()
args, kwargs = post.call_args
assert args[0] == "http://test-server:5002/desk/tts"
assert kwargs["headers"] == {"X-Desk-Api-Key": "test-key"}
assert kwargs["json"] == {
"session_id": "pet-test", "text": "hello there", "voice_id": "default-voice",
}
assert pcm.tolist() == [1, 2, 3]
assert rate == tts_config.TTS_SAMPLE_RATE
def test_synthesize_pcm_uses_the_override_voice():
with patch.object(tts.requests, "post") as post:
post.return_value = _mock_response(b"\x01\x00")
tts.synthesize_pcm("hi", voice_id="picked-voice")
assert post.call_args.kwargs["json"]["voice_id"] == "picked-voice"
def test_synthesize_pcm_raises_when_not_configured(monkeypatch):
monkeypatch.setattr(tts_config, "API_KEY", "")
with pytest.raises(TtsError, match="BOLT_SERVER_URL"):
tts.synthesize_pcm("hi")
def test_synthesize_pcm_raises_on_empty_audio():
with patch.object(tts.requests, "post") as post:
post.return_value = _mock_response(b"")
with pytest.raises(TtsError, match="no audio"):
tts.synthesize_pcm("hi")
def test_synthesize_pcm_raises_when_the_request_fails():
with patch.object(tts.requests, "post") as post:
post.return_value = _mock_response(b"", ok=False)
with pytest.raises(TtsError, match="server tts request failed"):
tts.synthesize_pcm("hi")
def test_stream_pcm_posts_to_the_same_endpoint_with_stream_true():
with patch.object(tts.requests, "post") as post:
response = MagicMock()
response.raise_for_status = MagicMock()
response.iter_content.return_value = [np.array([4, 5], dtype=np.int16).tobytes()]
post.return_value = response
chunks = list(tts.stream_pcm("hi"))
assert post.call_args.kwargs["stream"] is True
assert post.call_args[0][0] == "http://test-server:5002/desk/tts"
assert np.concatenate(chunks).tolist() == [4, 5]
def test_stream_pcm_raises_when_not_configured(monkeypatch):
monkeypatch.setattr(tts_config, "SERVER_URL", "")
with pytest.raises(TtsError, match="BOLT_SERVER_URL"):
list(tts.stream_pcm("hi"))