WildAi
ceea0c71cf
The upstream Chatterbox README documents paralinguistic tags ([laugh], [chuckle], [sigh], ...) as native to the Turbo family, and the vendored Turbo pack passes them straight to the tokenizer -- it never parses them. Our engine deleted them before the model saw them, so the feature looked implemented and was inaudible. * the text engine keeps a Turbo tag verbatim and records its name; on any other pack the nine Turbo names stay literal text instead of being swallowed as a voice name, and the run says so * the v2 angle tokens are re-spelled for Turbo (<giggle> -> [giggle]) and left untouched everywhere else * generate_audio now builds one plan for every family, synthesises one chunk per model call and joins the chunks; the classic path goes through the engine too, which is what makes "[Alice]" open a character segment * per-chunk seeds (seed + index) keep a chunked run reproducible without giving every chunk the same prosody * four appended widgets on ChatterboxTTS: enable_chunking, max_chars_per_chunk, chunk_combination_method, silence_between_chunks_ms * chunk boundaries prefer a sentence end and never land inside a tag * Turbo's exaggeration is passed as 0.0: the family ignores it for sampling and upstream warns about every non-zero value, once per chunk
556 lines
24 KiB
Python
556 lines
24 KiB
Python
"""S8.3 -- the tag engine: five features, in one order, that order load-bearing.
|
|
|
|
A tag engine that runs its passes in the wrong order does not crash. It quietly
|
|
produces different speech: ``[Alice] <giggle>`` loses its character if the
|
|
angle-bracket pass runs first, ``[de:Alice]`` becomes a voice called
|
|
``"de:Alice"`` if the character pass claims it, and ``[pause:1.5]`` says the
|
|
word "pause" out loud. None of that is catchable except by asserting the *order*,
|
|
which is why the ordering tests below are about which stage produced a value and
|
|
not only about what the final text says.
|
|
|
|
Every documented tag form is parsed here, the three interchangeable pause words
|
|
are parsed with each documented unit, unsupported per-segment parameters are
|
|
shown to be ignored rather than fatal, and ``auto``'s chunk-method selection is
|
|
pinned at every boundary the corpus names -- including 500, where the strict
|
|
comparison makes the answer surprising and therefore worth a test.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import pytest
|
|
|
|
from ComfyUI_Chatterbox.modules.text_processing import (
|
|
AUTO_CHUNK_THRESHOLDS,
|
|
CHUNK_COMBINATION_METHODS,
|
|
CRASH_PROTECTION_TEMPLATE,
|
|
MIN_SAFE_SEGMENT_CHARS,
|
|
PARAM_ALIASES,
|
|
TURBO_TAGS,
|
|
TURBO_TAG_V2_SPELLING,
|
|
V2_SPECIAL_TOKENS,
|
|
apply_crash_protection,
|
|
chunk_text,
|
|
parse_duration,
|
|
plan_text,
|
|
resolve_v2_tokens,
|
|
select_chunk_method,
|
|
split_segments,
|
|
)
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Durations and pause tags (stage 3)
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestDurations:
|
|
@pytest.mark.parametrize(
|
|
("value", "expected"),
|
|
[
|
|
("1.5", 1.5),
|
|
("2s", 2.0),
|
|
("500ms", 0.5),
|
|
("800MS", 0.8),
|
|
(" 1200 ms ", 1.2),
|
|
("3", 3.0),
|
|
("0.25s", 0.25),
|
|
(".5", 0.5),
|
|
],
|
|
)
|
|
def test_documented_spellings_parse(self, value: str, expected: float) -> None:
|
|
assert parse_duration(value) == pytest.approx(expected)
|
|
|
|
@pytest.mark.parametrize("value", ["", "abc", "5 parsecs", "-1", "1.2.3"])
|
|
def test_nonsense_is_refused(self, value: str) -> None:
|
|
"""A refusal is caught by the caller, which leaves the tag as literal text."""
|
|
with pytest.raises(ValueError):
|
|
parse_duration(value)
|
|
|
|
|
|
class TestPauseTagsAreInterchangeable:
|
|
"""``pause`` / ``wait`` / ``stop`` are three spellings of one tag."""
|
|
|
|
@pytest.mark.parametrize("word", ["pause", "wait", "stop"])
|
|
@pytest.mark.parametrize(
|
|
("value", "expected"),
|
|
[("1.5", 1.5), ("2s", 2.0), ("500ms", 0.5)],
|
|
)
|
|
def test_all_three_words_read_all_three_units(
|
|
self, word: str, value: str, expected: float
|
|
) -> None:
|
|
segments = split_segments(f"Hello there. [{word}:{value}] And more words.")
|
|
assert len(segments) == 2, "a pause opens a new segment"
|
|
assert segments[0].pause_after == pytest.approx(expected)
|
|
assert segments[0].text == "Hello there."
|
|
assert segments[1].text == "And more words."
|
|
assert segments[1].pause_after == 0.0
|
|
|
|
def test_the_word_is_case_insensitive(self) -> None:
|
|
assert split_segments("Hi. [PAUSE:2s] There.")[0].pause_after == 2.0
|
|
|
|
def test_an_unreadable_duration_is_kept_as_text_not_fatal(self) -> None:
|
|
segments = split_segments("Hi. [pause:soon] There.")
|
|
assert len(segments) == 1
|
|
assert "[pause:soon]" in segments[0].text
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Character switching (stage 1) and language tags (stage 2)
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestCharacterTags:
|
|
def test_a_bare_name_opens_a_segment(self) -> None:
|
|
segments = split_segments("[Alice] Hi there, Bob.")
|
|
assert len(segments) == 1
|
|
assert segments[0].character == "Alice"
|
|
assert segments[0].text == "Hi there, Bob."
|
|
|
|
def test_an_untagged_run_stays_in_the_same_segment(self) -> None:
|
|
assert len(split_segments("Just talking, with no tags at all.")) == 1
|
|
|
|
def test_two_characters_are_two_segments(self) -> None:
|
|
segments = split_segments("[Alice] One. [Bob] Two.")
|
|
assert [s.character for s in segments] == ["Alice", "Bob"]
|
|
assert [s.text for s in segments] == ["One.", "Two."]
|
|
|
|
def test_a_language_tag_is_not_a_character_named_de_alice(self) -> None:
|
|
"""The regression the ordering exists to prevent."""
|
|
segments = split_segments("[de:Alice] Hallo.")
|
|
assert len(segments) == 1
|
|
assert segments[0].character == "Alice"
|
|
assert segments[0].language == "de"
|
|
assert segments[0].text == "Hallo."
|
|
|
|
def test_a_display_name_resolves_to_its_code(self) -> None:
|
|
segments = split_segments("[German:Alice] Hallo.")
|
|
assert segments[0].language == "de"
|
|
assert segments[0].character == "Alice"
|
|
|
|
def test_a_language_tag_may_carry_no_character(self) -> None:
|
|
segments = split_segments("[de:] Hallo.")
|
|
assert segments[0].language == "de"
|
|
assert segments[0].character is None
|
|
|
|
def test_a_pause_tag_is_not_a_character_named_pause(self) -> None:
|
|
"""It is a pause, and a leading one has nothing to be silent after."""
|
|
segments = split_segments("[pause:1.5] Hallo.")
|
|
assert len(segments) == 1
|
|
assert segments[0].character is None
|
|
assert segments[0].text == "Hallo."
|
|
assert segments[0].pause_after == 0.0, (
|
|
"a pause with no text before it has no segment to attach to"
|
|
)
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Per-segment overrides (stage 4)
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestPerSegmentOverrides:
|
|
def test_order_does_not_matter(self) -> None:
|
|
first = split_segments("[Alice|seed:42|temp:0.5] Hi.")[0]
|
|
second = split_segments("[Alice|temp:0.5|seed:42] Hi.")[0]
|
|
assert first.params == second.params
|
|
assert first.params == {"seed": "42", "temperature": "0.5"}
|
|
|
|
def test_case_does_not_matter(self) -> None:
|
|
segment = split_segments("[Alice|SEED:42|Temp:0.5] Hi.")[0]
|
|
assert segment.params == {"seed": "42", "temperature": "0.5"}
|
|
|
|
@pytest.mark.parametrize(
|
|
("alias", "canonical"),
|
|
[("temp", "temperature"), ("cfg", "cfg_weight"),
|
|
("cfg_weight", "cfg_weight"), ("exag", "exaggeration")],
|
|
)
|
|
def test_documented_aliases_resolve(self, alias: str, canonical: str) -> None:
|
|
segment = split_segments(f"[Alice|{alias}:0.5] Hi.")[0]
|
|
assert segment.params == {canonical: "0.5"}
|
|
assert canonical in PARAM_ALIASES.values()
|
|
|
|
def test_an_unsupported_parameter_is_ignored_not_fatal(self) -> None:
|
|
"""A script carried over from another tool must still run."""
|
|
segment = split_segments("[Alice|batch:4|temp:0.5] Hi.")[0]
|
|
assert segment.params == {"temperature": "0.5"}
|
|
assert segment.character == "Alice"
|
|
assert segment.text == "Hi."
|
|
|
|
def test_a_parameter_with_no_socket_at_all_is_ignored(self) -> None:
|
|
segment = split_segments("[Alice|batch_size:8] Hi.")[0]
|
|
assert segment.params == {}
|
|
assert segment.text == "Hi."
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# v2 angle-bracket tokens (stage 5) -- and the order it runs in
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestV2Tokens:
|
|
def test_all_thirty_documented_names_are_known(self) -> None:
|
|
assert len(V2_SPECIAL_TOKENS) == 30
|
|
assert len(set(V2_SPECIAL_TOKENS)) == 30
|
|
|
|
@pytest.mark.parametrize("token", V2_SPECIAL_TOKENS)
|
|
def test_every_documented_token_is_recognised(self, token: str) -> None:
|
|
resolved, seen = resolve_v2_tokens(f"Hello <{token}> there.")
|
|
assert seen == (token,), f"<{token}> was not recognised"
|
|
assert resolved == f"Hello <{token}> there."
|
|
|
|
def test_casing_is_normalised_except_where_it_is_significant(self) -> None:
|
|
assert resolve_v2_tokens("<GIGGLE>")[0] == "<giggle>"
|
|
assert resolve_v2_tokens("<UH>")[0] == "<UH>"
|
|
assert resolve_v2_tokens("<um>")[0] == "<UM>"
|
|
|
|
def test_an_unknown_angle_token_is_left_alone(self) -> None:
|
|
"""The upstream tokenizer carries 118 special tokens; only 30 are named.
|
|
|
|
Deleting a word the model can still speak is worse than passing it on.
|
|
"""
|
|
resolved, seen = resolve_v2_tokens("Hello <not_a_real_token> there.")
|
|
assert seen == ()
|
|
assert resolved == "Hello <not_a_real_token> there."
|
|
|
|
def test_a_giggle_after_a_character_resolves_after_the_character(self) -> None:
|
|
"""The load-bearing ordering assertion.
|
|
|
|
Stage 5 must not have run yet when the character is resolved, or
|
|
``[Alice] <giggle>`` loses its voice.
|
|
"""
|
|
segments = split_segments("[Alice] <giggle> Hello there, nice to meet you.")
|
|
assert len(segments) == 1
|
|
assert segments[0].character == "Alice", (
|
|
"the character was not resolved; stage 5 ran before stage 1"
|
|
)
|
|
assert "<giggle>" in segments[0].text, (
|
|
"the v2 token was rewritten during the character stage; it must still "
|
|
"be untouched text when the character resolves"
|
|
)
|
|
|
|
plan = plan_text("[Alice] <giggle> Hello there, nice to meet you.")
|
|
assert plan.segments[0].character == "Alice"
|
|
assert "<giggle>" in plan.text
|
|
assert "[" not in plan.text
|
|
|
|
def test_the_same_example_composes_with_a_pause(self) -> None:
|
|
"""The corpus's own worked example, end to end."""
|
|
plan = plan_text(
|
|
"[Alice] Hello! <giggle> Nice to meet you. [pause:0.5] How are you? "
|
|
"<whisper> I have a secret."
|
|
)
|
|
assert [s.character for s in plan.segments] == ["Alice", None]
|
|
assert plan.segments[0].pause_after == pytest.approx(0.5)
|
|
assert plan.segments[1].pause_after == 0.0
|
|
assert "<giggle>" in plan.text
|
|
assert "<whisper>" in plan.text
|
|
|
|
|
|
class TestV2TokensOnTheTurboPath:
|
|
"""A Turbo checkpoint knows the square spelling and not the angle one.
|
|
|
|
The v2 set writes ``<clear_throat>`` with an underscore and Turbo's
|
|
tokenizer was trained on ``[clear throat]`` with a space, so the one
|
|
spelling that differs is published rather than buried.
|
|
"""
|
|
|
|
def test_v2_tokens_are_untouched_on_the_classic_path(self) -> None:
|
|
assert resolve_v2_tokens("<giggle>")[0] == "<giggle>"
|
|
|
|
def test_v2_tokens_become_square_brackets_on_the_turbo_path(self) -> None:
|
|
assert resolve_v2_tokens("<giggle>", turbo_tags=True)[0] == "[giggle]"
|
|
|
|
def test_clear_throat_uses_the_turbo_spelling(self) -> None:
|
|
assert resolve_v2_tokens("<clear_throat>", turbo_tags=True)[0] == (
|
|
"[clear throat]"
|
|
)
|
|
|
|
def test_uh_keeps_its_casing_in_both_spellings(self) -> None:
|
|
assert resolve_v2_tokens("<UH>")[0] == "<UH>"
|
|
assert resolve_v2_tokens("<UH>", turbo_tags=True)[0] == "[UH]"
|
|
|
|
def test_an_unknown_angle_token_is_left_alone_on_the_turbo_path(self) -> None:
|
|
resolved, seen = resolve_v2_tokens("<not_a_real_token>", turbo_tags=True)
|
|
assert seen == ()
|
|
assert resolved == "<not_a_real_token>"
|
|
|
|
def test_a_v2_token_after_a_character_keeps_the_character(self) -> None:
|
|
"""The conversion happens after the character is resolved, so the
|
|
character survives it."""
|
|
segments = split_segments("[Alice] <giggle> Hello.", turbo_tags=True)
|
|
assert segments[0].character == "Alice"
|
|
assert "<giggle>" in segments[0].text
|
|
|
|
plan = plan_text("[Alice] <giggle> Hello.", turbo_tags=True)
|
|
assert plan.segments[0].character == "Alice"
|
|
assert "[giggle]" in plan.text
|
|
|
|
def test_the_turbo_plan_text_carries_the_square_spelling(self) -> None:
|
|
# Crash protection is off here: it pads any segment shorter than
|
|
# twenty-one characters, and a two-word script is shorter than that.
|
|
plan = plan_text("<giggle> Hi.", turbo_tags=True, crash_protection_template="")
|
|
assert plan.text == "[giggle] Hi."
|
|
|
|
def test_the_only_differing_spelling_is_published(self) -> None:
|
|
assert TURBO_TAG_V2_SPELLING == {"clear_throat": "clear throat"}
|
|
|
|
|
|
class TestTurboTagsAreADifferentVocabulary:
|
|
"""``[laugh]`` is a Turbo tag, and a Turbo tag is never eaten."""
|
|
|
|
def test_a_turbo_tag_is_preserved_in_the_segment_text(self) -> None:
|
|
"""The tag reaches the model exactly as the user typed it.
|
|
|
|
Upstream does not parse these tags; it hands the text to the
|
|
tokenizer, which knows the square spelling. A tag that is deleted
|
|
before the model sees it is a feature that looks implemented and is
|
|
not.
|
|
"""
|
|
segment = split_segments("[laugh] Hi.", turbo_tags=True)[0]
|
|
assert segment.text == "[laugh] Hi."
|
|
|
|
def test_a_turbo_tag_is_still_recorded(self) -> None:
|
|
"""Recognition is kept; only the deletion is gone."""
|
|
segment = split_segments("[laugh] Hi.", turbo_tags=True)[0]
|
|
assert segment.turbo_tags == ("laugh",)
|
|
assert segment.character is None
|
|
|
|
def test_a_turbo_tag_at_the_start_of_a_segment_is_preserved(self) -> None:
|
|
segment = split_segments("[Alice] [chuckle] Hello.", turbo_tags=True)
|
|
assert segment[0].text == "[chuckle] Hello."
|
|
assert segment[0].turbo_tags == ("chuckle",)
|
|
assert segment[0].character == "Alice"
|
|
|
|
def test_two_turbo_tags_are_both_preserved_and_recorded(self) -> None:
|
|
segment = split_segments("[laugh] and [chuckle] that", turbo_tags=True)[0]
|
|
assert segment.text == "[laugh] and [chuckle] that"
|
|
assert segment.turbo_tags == ("laugh", "chuckle")
|
|
|
|
def test_the_same_text_is_a_tag_on_the_turbo_path(self) -> None:
|
|
segments = split_segments("[laugh] Hello.", turbo_tags=True)
|
|
assert segments[0].character is None
|
|
assert segments[0].turbo_tags == ("laugh",)
|
|
|
|
def test_turbo_tags_are_recognised_case_insensitively(self) -> None:
|
|
assert split_segments("[Clear Throat] Hi.", turbo_tags=True)[0].turbo_tags == (
|
|
"clear throat",
|
|
)
|
|
|
|
def test_the_turbo_set_is_not_the_v2_set(self) -> None:
|
|
"""Nine square-bracket tokens against thirty angle-bracket ones.
|
|
|
|
Five names are shared between the two vocabularies (``sigh``, ``gasp``,
|
|
``cough``, ``sniff``, ``groan``); what differs is the *syntax* each
|
|
vocabulary uses, which is the point of this module.
|
|
"""
|
|
assert len(TURBO_TAGS) == 9
|
|
assert len(V2_SPECIAL_TOKENS) == 30
|
|
shared = set(TURBO_TAGS) & {token.lower() for token in V2_SPECIAL_TOKENS}
|
|
assert shared == {"sigh", "gasp", "cough", "sniff", "groan"}
|
|
assert "clear throat" not in V2_SPECIAL_TOKENS
|
|
assert "clear_throat" not in TURBO_TAGS
|
|
|
|
def test_an_angle_token_is_still_normalised_on_the_turbo_path(self) -> None:
|
|
segments = split_segments("[laugh] <giggle> Hi.", turbo_tags=True)
|
|
assert segments[0].turbo_tags == ("laugh",)
|
|
assert "<giggle>" in segments[0].text
|
|
|
|
|
|
class TestATurboTagIsNotAVoiceName:
|
|
"""On a pack that cannot speak them, Turbo tags are text -- not a name.
|
|
|
|
Treating ``[laugh]`` as a voice called "laugh" swallows the word: the
|
|
script says nothing where the user wrote a direction, and the voice never
|
|
changes either, because this repository ships no voice library.
|
|
"""
|
|
|
|
def test_a_turbo_tag_name_on_the_classic_path_is_not_a_character(self) -> None:
|
|
segment = split_segments("[laugh] Hello.")[0]
|
|
assert segment.character is None
|
|
assert "[laugh]" in segment.text
|
|
|
|
def test_an_unknown_bracket_on_the_classic_path_is_still_a_character(self) -> None:
|
|
"""The unchanged behaviour: every other bare bracket is a name."""
|
|
segment = split_segments("[Alice] Hello.")[0]
|
|
assert segment.character == "Alice"
|
|
assert segment.text == "Hello."
|
|
|
|
def test_a_turbo_tag_name_on_the_classic_path_produces_a_notice(self) -> None:
|
|
notices = plan_text("[laugh] Hi.").notices
|
|
assert notices, "the tag was swallowed without a word"
|
|
assert any("Turbo" in notice and "laugh" in notice for notice in notices)
|
|
|
|
def test_the_classic_path_does_not_record_the_tag_as_a_turbo_tag(self) -> None:
|
|
"""It is not a Turbo tag here -- it is the word the user typed."""
|
|
assert split_segments("[laugh] Hello.")[0].turbo_tags == ()
|
|
|
|
@pytest.mark.parametrize("body", ["de:Alice", "Alice|seed:42", "Alice|temp:0.5"])
|
|
def test_a_bracket_with_a_colon_or_pipe_is_never_treated_as_a_turbo_tag(
|
|
self, body: str
|
|
) -> None:
|
|
"""Only the nine bare names are Turbo tags; the other forms keep
|
|
their own meaning."""
|
|
segment = split_segments(f"[{body}] Hello.")[0]
|
|
assert segment.turbo_tags == ()
|
|
|
|
def test_a_turbo_tag_is_not_noticed_on_the_turbo_path(self) -> None:
|
|
"""It is native there, so there is nothing to announce."""
|
|
plan = plan_text("[laugh] Hi.", turbo_tags=True)
|
|
assert not [notice for notice in plan.notices if "Turbo" in notice]
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Chunking
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestAutoChunkMethodBoundaries:
|
|
"""``auto`` selects by length, and both comparisons are strict.
|
|
|
|
The corpus phrases the rule as "<500 -> concatenate" while the thresholds are
|
|
written as ">1000" and ">500". Those disagree at exactly 500, so the strict
|
|
reading is implemented and 500 is pinned here -- a rule with an unstated
|
|
boundary is a rule with a bug in it.
|
|
"""
|
|
|
|
@pytest.mark.parametrize(
|
|
("chars", "expected"),
|
|
[
|
|
(499, "concatenate"),
|
|
(500, "concatenate"),
|
|
(501, "crossfade"),
|
|
(1000, "crossfade"),
|
|
(1001, "silence_padding"),
|
|
(5000, "silence_padding"),
|
|
],
|
|
)
|
|
def test_the_documented_boundaries(self, chars: int, expected: str) -> None:
|
|
assert select_chunk_method("auto", chars) == expected
|
|
assert plan_text("x" * chars).chunk_method == expected
|
|
|
|
def test_the_thresholds_are_published_not_implicit(self) -> None:
|
|
assert AUTO_CHUNK_THRESHOLDS == ((1000, "silence_padding"), (500, "crossfade"))
|
|
|
|
@pytest.mark.parametrize(
|
|
"method", [name for name in CHUNK_COMBINATION_METHODS if name != "auto"]
|
|
)
|
|
def test_an_explicit_method_passes_through(self, method: str) -> None:
|
|
"""Whatever the user picked is what runs, at any length.
|
|
|
|
``auto`` is the only value that gets *replaced*; the other three are
|
|
names, not hints, and a short script must not quietly switch the user
|
|
off ``silence_padding`` because it was under a thousand characters.
|
|
"""
|
|
assert select_chunk_method(method, 10) == method
|
|
assert select_chunk_method(method, 5000) == method
|
|
|
|
def test_the_resolved_method_is_never_auto(self) -> None:
|
|
for chars in (0, 1, 500, 5000):
|
|
assert select_chunk_method("auto", chars) != "auto"
|
|
|
|
|
|
class TestChunking:
|
|
def test_a_short_script_is_one_chunk(self) -> None:
|
|
assert chunk_text("Short enough.", max_chars_per_chunk=400) == ["Short enough."]
|
|
|
|
def test_a_long_script_is_cut(self) -> None:
|
|
text = "A sentence here. " * 100
|
|
chunks = chunk_text(text, 200)
|
|
assert len(chunks) > 1
|
|
assert all(len(chunk) <= 200 for chunk in chunks)
|
|
assert "".join(chunks).replace(" ", "") == text.replace(" ", "")
|
|
# No word is lost or split in half on the way through.
|
|
assert text.split() == " ".join(chunks).split()
|
|
|
|
def test_cutting_is_disabled_by_the_flag(self) -> None:
|
|
text = "A sentence here. " * 100
|
|
assert chunk_text(text, 200, enabled=False) == [text]
|
|
|
|
def test_a_crash_protected_short_script_is_still_short(self) -> None:
|
|
plan = plan_text("Hi", enable_chunking=True, max_chars_per_chunk=400)
|
|
assert len(plan.chunks) == 1
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# The whole pipeline
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestPlanText:
|
|
def test_a_plain_script_yields_one_segment_and_one_chunk(self) -> None:
|
|
plan = plan_text("Hello there, this is a plain script.")
|
|
assert len(plan.segments) == 1
|
|
assert plan.chunks == ("Hello there, this is a plain script.",)
|
|
assert plan.char_count == len(plan.text)
|
|
|
|
def test_blank_input_is_an_empty_plan_not_an_error(self) -> None:
|
|
plan = plan_text(" ")
|
|
assert plan.segments == ()
|
|
assert plan.chunks == ("",)
|
|
|
|
def test_characters_are_listed_in_first_appearance_order(self) -> None:
|
|
plan = plan_text("[Bob] One. [Alice] Two. [Bob] Three.")
|
|
assert plan.characters == ("Bob", "Alice")
|
|
|
|
def test_the_default_template_is_the_documented_one(self) -> None:
|
|
assert CRASH_PROTECTION_TEMPLATE == "hmm ,, {seg} hmm ,,"
|
|
assert MIN_SAFE_SEGMENT_CHARS == 21
|
|
assert apply_crash_protection("Short.").startswith("hmm ,,")
|
|
|
|
def test_auto_is_unchanged_by_the_new_boundary_rule(self) -> None:
|
|
"""The tag-safe cut changes *where* a script is cut, not *how long* it
|
|
is: ``auto`` still resolves on the script's length alone."""
|
|
assert plan_text("x" * 500).chunk_method == "concatenate"
|
|
assert plan_text("x" * 501).chunk_method == "crossfade"
|
|
assert plan_text("x" * 1001).chunk_method == "silence_padding"
|
|
|
|
def test_a_script_with_tags_resolves_auto_on_its_length(self) -> None:
|
|
script = "[laugh] " + ("x" * 600)
|
|
assert plan_text(script).chunk_method == "crossfade"
|
|
|
|
|
|
# ---------------------------------------------------------------------------
|
|
# Notices -- what the engine could not honour
|
|
# ---------------------------------------------------------------------------
|
|
|
|
|
|
class TestNotices:
|
|
"""A tag that is parsed and then means nothing is announced, not dropped.
|
|
|
|
The alternative is a script that quietly loses a word, and a user who has
|
|
no idea why ``[laugh]`` did nothing.
|
|
"""
|
|
|
|
def test_a_plain_script_has_no_notices(self) -> None:
|
|
assert plan_text("Hello there, this is a plain script.").notices == ()
|
|
|
|
def test_a_turbo_tag_on_a_non_turbo_pack_is_noticed(self) -> None:
|
|
notices = plan_text("[laugh] Hi.").notices
|
|
assert len(notices) == 1
|
|
assert "[laugh]" in notices[0]
|
|
assert "Turbo" in notices[0]
|
|
|
|
def test_the_same_tag_is_not_noticed_twice(self) -> None:
|
|
plan = plan_text("[laugh] Hi. [laugh] Bye.")
|
|
assert len([notice for notice in plan.notices if "Turbo" in notice]) == 1
|
|
|
|
def test_a_pause_is_noticed_rather_than_silently_dropped(self) -> None:
|
|
notices = plan_text("Hi there. [pause:1.5] And more.").notices
|
|
assert any("pause" in notice.lower() for notice in notices)
|
|
|
|
def test_a_language_is_noticed_rather_than_silently_dropped(self) -> None:
|
|
notices = plan_text("[de:Alice] Hallo.").notices
|
|
assert any("language" in notice.lower() for notice in notices)
|
|
|
|
def test_a_notice_never_changes_the_audio(self) -> None:
|
|
"""A notice is a note about the text, not a change to it: the same
|
|
words are spoken on either path."""
|
|
classic = plan_text("[laugh] Hi there, this is long enough to speak.")
|
|
turbo = plan_text(
|
|
"[laugh] Hi there, this is long enough to speak.", turbo_tags=True
|
|
)
|
|
assert classic.text == turbo.text
|
|
assert classic.chunks == turbo.chunks
|
|
assert classic.notices != ()
|
|
assert turbo.notices == ()
|