huri-module-tag 0.1.0__tar.gz

huri_module_tag-0.1.0/PKG-INFO

@@ -0,0 +1,36 @@
+Metadata-Version: 2.4
+Name: huri-module-tag
+Version: 0.1.0
+Summary: HuRI module: Text aggregator: stitches overlapping transcripts into one question
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: huri-core[server]<0.5,>=0.4
+Requires-Dist: huri-events<0.2,>=0.1
+
+# huri-module-tag
+
+Text aggregator: stitches overlapping transcripts into one question.
+
+| Module | Topics | What it does |
+|---|---|---|
+| `tag` | `huri.transcript` → `huri.partial_question` | Stitches overlapping transcript passes into one sentence |
+
+```sh
+pip install -e packages/huri-module-tag     # from a huri-modules checkout, next to huri-core and huri-events
+HURI_MODULES=tag ...   # enable it on the server
+```
+
+A distribution of
+[huri-modules](https://github.com/Lucy-Robotics/huri-modules), versioned in its
+`pyproject.toml` (see RELEASING.md there). It depends on `huri-core` and
+`huri-events`, which are not on PyPI: without a checkout, install them from git
+in the same pip command, so pip never looks them up on an index:
+
+```sh
+ref=0.1.0   # a huri-modules release tag
+core=0.4.0  # HURI_CORE_REF in that release's versions.env
+pip install \
+  "huri-core[server] @ git+https://github.com/Lucy-Robotics/huri-core.git@$core" \
+  "huri-events @ git+https://github.com/Lucy-Robotics/huri-modules.git@$ref#subdirectory=packages/huri-events" \
+  "huri-module-tag @ git+https://github.com/Lucy-Robotics/huri-modules.git@$ref#subdirectory=packages/huri-module-tag"
+```

huri_module_tag-0.1.0/README.md

@@ -0,0 +1,27 @@
+# huri-module-tag
+
+Text aggregator: stitches overlapping transcripts into one question.
+
+| Module | Topics | What it does |
+|---|---|---|
+| `tag` | `huri.transcript` → `huri.partial_question` | Stitches overlapping transcript passes into one sentence |
+
+```sh
+pip install -e packages/huri-module-tag     # from a huri-modules checkout, next to huri-core and huri-events
+HURI_MODULES=tag ...   # enable it on the server
+```
+
+A distribution of
+[huri-modules](https://github.com/Lucy-Robotics/huri-modules), versioned in its
+`pyproject.toml` (see RELEASING.md there). It depends on `huri-core` and
+`huri-events`, which are not on PyPI: without a checkout, install them from git
+in the same pip command, so pip never looks them up on an index:
+
+```sh
+ref=0.1.0   # a huri-modules release tag
+core=0.4.0  # HURI_CORE_REF in that release's versions.env
+pip install \
+  "huri-core[server] @ git+https://github.com/Lucy-Robotics/huri-core.git@$core" \
+  "huri-events @ git+https://github.com/Lucy-Robotics/huri-modules.git@$ref#subdirectory=packages/huri-events" \
+  "huri-module-tag @ git+https://github.com/Lucy-Robotics/huri-modules.git@$ref#subdirectory=packages/huri-module-tag"
+```

huri_module_tag-0.1.0/huri_module_tag.egg-info/PKG-INFO

@@ -0,0 +1,36 @@
+Metadata-Version: 2.4
+Name: huri-module-tag
+Version: 0.1.0
+Summary: HuRI module: Text aggregator: stitches overlapping transcripts into one question
+Requires-Python: >=3.10
+Description-Content-Type: text/markdown
+Requires-Dist: huri-core[server]<0.5,>=0.4
+Requires-Dist: huri-events<0.2,>=0.1
+
+# huri-module-tag
+
+Text aggregator: stitches overlapping transcripts into one question.
+
+| Module | Topics | What it does |
+|---|---|---|
+| `tag` | `huri.transcript` → `huri.partial_question` | Stitches overlapping transcript passes into one sentence |
+
+```sh
+pip install -e packages/huri-module-tag     # from a huri-modules checkout, next to huri-core and huri-events
+HURI_MODULES=tag ...   # enable it on the server
+```
+
+A distribution of
+[huri-modules](https://github.com/Lucy-Robotics/huri-modules), versioned in its
+`pyproject.toml` (see RELEASING.md there). It depends on `huri-core` and
+`huri-events`, which are not on PyPI: without a checkout, install them from git
+in the same pip command, so pip never looks them up on an index:
+
+```sh
+ref=0.1.0   # a huri-modules release tag
+core=0.4.0  # HURI_CORE_REF in that release's versions.env
+pip install \
+  "huri-core[server] @ git+https://github.com/Lucy-Robotics/huri-core.git@$core" \
+  "huri-events @ git+https://github.com/Lucy-Robotics/huri-modules.git@$ref#subdirectory=packages/huri-events" \
+  "huri-module-tag @ git+https://github.com/Lucy-Robotics/huri-modules.git@$ref#subdirectory=packages/huri-module-tag"
+```

huri_module_tag-0.1.0/huri_module_tag.egg-info/SOURCES.txt

@@ -0,0 +1,12 @@
+README.md
+pyproject.toml
+huri_module_tag.egg-info/PKG-INFO
+huri_module_tag.egg-info/SOURCES.txt
+huri_module_tag.egg-info/dependency_links.txt
+huri_module_tag.egg-info/entry_points.txt
+huri_module_tag.egg-info/requires.txt
+huri_module_tag.egg-info/top_level.txt
+huri_tag/__init__.py
+huri_tag/huri_module.toml
+huri_tag/module.py
+tests/test_tag.py

huri_module_tag-0.1.0/huri_module_tag.egg-info/dependency_links.txt

@@ -0,0 +1 @@
+

huri_module_tag-0.1.0/huri_module_tag.egg-info/entry_points.txt

@@ -0,0 +1,2 @@
+[huri.modules]
+tag = huri_tag.module:TAG

huri_module_tag-0.1.0/huri_module_tag.egg-info/requires.txt

@@ -0,0 +1,2 @@
+huri-core[server]<0.5,>=0.4
+huri-events<0.2,>=0.1

huri_module_tag-0.1.0/huri_module_tag.egg-info/top_level.txt

@@ -0,0 +1 @@
+huri_tag

huri_module_tag-0.1.0/huri_tag/huri_module.toml

@@ -0,0 +1,7 @@
+name = "tag"
+summary = "Stitches overlapping transcript passes into one sentence"
+input = { topic = "huri.transcript", event = "huri_events.speech:Transcript" }
+output = { topic = "huri.partial_question", event = "huri_events.question:PartialQuestion" }
+events = ["huri-events>=0.1,<0.2"]
+requires = { gpu = false, actuation = false }
+huri_core = ">=0.4,<0.5"

huri_module_tag-0.1.0/huri_tag/module.py

@@ -0,0 +1,89 @@
+import re
+from typing import Optional
+
+from huri_events.question import PartialQuestion
+from huri_events.speech import Transcript
+
+from huri_core.core.module import Module
+
+
+class TAG(Module):
+    """TAG Module
+
+    Aggregate all transcriptions and send when transcript end.
+
+    Whisper transcribes each sliding window independently, so the start of each
+    transcript repeats the end of the previous one (with possibly different
+    casing or punctuation). Transcripts are stitched together on that overlap.
+
+    input: huri.transcript,
+    output: huri.partial_question
+
+    :min_prefix: shortest word fragment (in characters) that may match a longer
+        word when merging overlapping transcripts, e.g. "weath" and "weather".
+        Guards against 1-2 letter fragments matching unrelated words.
+    """
+
+    input_type = "huri.transcript"
+    output_type = "huri.partial_question"
+    input_event = Transcript
+    output_event = PartialQuestion
+
+    def __init__(
+        self,
+        min_prefix: int = 3,
+    ):
+        super().__init__()
+
+        self.min_prefix = min_prefix
+
+        self.sentence: str = ""
+
+    @staticmethod
+    def _norm(word: str) -> str:
+        return re.sub(r"[^\w]", "", word.lower())
+
+    def _same_word(self, a: str, b: str) -> bool:
+        """Words match after normalisation, or when one is a truncated form of
+        the other (a window edge can cut a word in half: "weath" / "weather")."""
+
+        na, nb = self._norm(a), self._norm(b)
+        if na == nb:
+            return True
+        short, long_ = sorted((na, nb), key=len)
+        return len(short) >= self.min_prefix and long_.startswith(short)
+
+    def _merge(self, current: str, new: str) -> str:
+        """Find the longest run of words where the tail of ``current`` equals the
+        head of ``new`` and append only what follows it. When there is no
+        overlap, ``new`` is appended whole: text is never dropped from either
+        side.
+        """
+
+        cur_words = current.split()
+        new_words = new.split()
+        if not cur_words:
+            return new.strip()
+        if not new_words:
+            return current.strip()
+
+        for k in range(min(len(cur_words), len(new_words)), 0, -1):
+            if all(
+                self._same_word(a, b) for a, b in zip(cur_words[-k:], new_words[:k])
+            ):
+                # Keep the newer spelling of the overlap (more audio context).
+                return " ".join(cur_words[:-k] + new_words)
+        return " ".join(cur_words + new_words)
+
+    async def process(self, transcript: Transcript) -> Optional[PartialQuestion]:
+        text = transcript.text
+
+        if text != "":
+            self.sentence = self._merge(self.sentence, text)
+
+        if transcript.end and self.sentence != "":
+            transcript = Transcript(self.sentence, True)
+            self.sentence = ""
+            return PartialQuestion(transcript=transcript, emotion=None)
+        else:
+            return None

huri_module_tag-0.1.0/pyproject.toml

@@ -0,0 +1,23 @@
+[build-system]
+requires = ["setuptools>=64,<82"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "huri-module-tag"
+version = "0.1.0"
+description = "HuRI module: Text aggregator: stitches overlapping transcripts into one question"
+readme = "README.md"
+requires-python = ">=3.10"
+dependencies = [
+  "huri-core[server]>=0.4,<0.5",
+  "huri-events>=0.1,<0.2",
+]
+
+[project.entry-points."huri.modules"]
+tag = "huri_tag.module:TAG"
+
+[tool.setuptools.packages.find]
+include = ["huri_tag*"]
+
+[tool.setuptools.package-data]
+huri_tag = ["huri_module.toml", "manifests/*/huri_module.toml"]

huri_module_tag-0.1.0/setup.cfg

@@ -0,0 +1,4 @@
+[egg_info]
+tag_build = 
+tag_date = 0
+

huri_module_tag-0.1.0/tests/test_tag.py

@@ -0,0 +1,83 @@
+import asyncio
+
+import pytest
+from huri_events.speech import Transcript
+from huri_tag.module import TAG
+
+from huri_core.testing import ModuleHarness
+
+
+@pytest.mark.parametrize(
+    "current,new,expected",
+    [
+        (
+            "what is the weather",
+            "the weather like in Paris today",
+            "what is the weather like in Paris today",
+        ),
+        # No overlap must never corrupt/replace the sentence.
+        (
+            "what is the weather like",
+            "in Paris today",
+            "what is the weather like in Paris today",
+        ),
+        (
+            "What is the weather like.",
+            "weather like in Paris today?",
+            "What is the weather like in Paris today?",
+        ),
+        (
+            "I want to book a table for two",
+            "for two people at eight",
+            "I want to book a table for two people at eight",
+        ),
+        ("it is a very very", "very very long day", "it is a very very long day"),
+        # Word cut by the window edge.
+        ("what is the weath", "weather like", "what is the weather like"),
+        # New text fully contained in the tail.
+        ("hello there my friend", "my friend", "hello there my friend"),
+        ("", "hello", "hello"),
+        ("hello", "", "hello"),
+    ],
+)
+def test_merge(current, new, expected):
+    assert TAG()._merge(current, new) == expected
+
+
+def test_tag_emits_on_end_and_resets():
+    out = ModuleHarness(TAG).run(
+        Transcript("what is the", False),
+        Transcript("the weather", False),
+        Transcript("weather like", True),
+        Transcript("", True),
+        Transcript("next question", True),
+    )
+    assert [q.transcript.text for q in out] == [
+        "what is the weather like",
+        "next question",
+    ]
+
+
+def test_stitches_the_overlapping_passes_stt_emits():
+    out = ModuleHarness(TAG).run(
+        Transcript("what is the", False), Transcript("the weather", True)
+    )
+    assert [q.transcript.text for q in out] == ["what is the weather"]
+
+
+def test_min_prefix_is_configurable():
+    # "th" (from a cut "the") only matches when min_prefix allows 2 chars.
+    assert TAG(min_prefix=3)._merge("what is th", "the weather") == (
+        "what is th the weather"
+    )
+    assert TAG(min_prefix=2)._merge("what is th", "the weather") == (
+        "what is the weather"
+    )
+
+    async def run():
+        tag = TAG(min_prefix=2)
+        await tag.process(Transcript("what is th", False))
+        q = await tag.process(Transcript("the weather", True))
+        assert q.transcript.text == "what is the weather"
+
+    asyncio.run(run())