speechmetryflow 0.2.2__py3-none-any.whl

speechmetryflow/semantic/assets.py

@@ -0,0 +1,285 @@
+# ICU dictionaries for both languages
+ICU_cookie_fr = {
+    "mère": ["mère", "maman", "mamie", "mam", "mamma", "momma", "ma", "mama", "femme"],
+    "garçon": [
+        "garçon",
+        "gamin",
+        "mec",
+        "gosse",
+        "fils",
+        "écolier",
+        "jeune homme",
+        "jeune garçon",
+        "enfant mâle",
+    ],
+    "fille": [
+        "fille",
+        "fille",
+        "jeune fille",
+        "écolière",
+        "jeune dame",
+        "jeune fille",
+        "enfant femelle",
+    ],
+    "cuisine": [
+        "cuisine",
+    ],
+    "extérieur": ["extérieur", "dehors", "à l'air libre", "en plein air"],
+    "robinet": ["robinet", "évacuation", "mitigeur"],
+    "eau": [
+        "eau",
+    ],
+    "évier": [
+        "évier",
+    ],
+    "sol": ["sol", "terre", "plancher"],
+    "assiette": ["assiette", "plat", "plateau", "soucoupe"],
+    "vaisselle_sur_comptoir": ["posée sur", "placée sur", "étalée sur"],
+    "comptoir": ["comptoir", "plan de travail"],
+    "cookies": ["cookies", "biscuit", "biscuits"],
+    "bocal": ["bocal", "pot", "boîte"],
+    "placard": ["placard", "armoire", "placards", "armoires"],
+    "tabouret": ["tabouret", "escabeau"],
+    "fenêtre": [
+        "fenêtre",
+    ],
+    "rideau": ["rideau", "store", "écran", "volet"],
+    "garçon_prenant_cookie": [
+        "prenant",
+        "saisissant",
+        "atteignant",
+        "prendre",
+        "saisir",
+        "atteindre",
+        "se saisir",
+    ],
+    "garçon_tabouret_tombant": ["tombant", "basculant", "tomber", "basculer"],
+    "mère_séchant_lavant_vaisselle": [
+        "sécher",
+        "séchant",
+        "laver",
+        "lavant",
+        "nettoyer",
+        "nettoyant",
+        "frotter",
+        "frottant",
+        "essuyant",
+        "faire la vaisselle",
+    ],
+    "eau_débordant": [
+        "déborder",
+        "débordant",
+        "renverser",
+        "renversant",
+        "cascader",
+        "évier débordant",
+        "débordement",
+    ],
+    "fille_demandant_cookie": [
+        "demander",
+        "demandant",
+        "vouloir",
+        "veut",
+        "solliciter",
+        "sollicite",
+    ],
+    "mère_indifférente_au_débordement": [
+        "indifférente",
+        "pas dérangée",
+        "se soucier",
+        "pas dérangé",
+        "remarqué",
+        "déranger",
+    ],
+    "mère_ne_remarquant_pas_les_enfants": [
+        "inconsciente",
+        "remarqué",
+        "pas remarqué",
+        "pas prendre garde",
+        "pas dérangée",
+        "pas dérangé",
+    ],
+}
+
+ICU_cookie_en = {
+    "mother": [
+        "mother",
+        "mom",
+        "mum",
+        "mommy",
+        "mummy",
+        "mamma",
+        "momma",
+        "ma",
+        "mama",
+        "woman",
+    ],
+    "boy": [
+        "boy",
+        "guy",
+        "dude",
+        "lad",
+        "son",
+        "schoolboy",
+        "young man",
+        "son",
+        "young boy",
+        "male child",
+    ],
+    "girl": [
+        "girl",
+        "daughter",
+        "lass",
+        "schoolgirl",
+        "young lady",
+        "young girl",
+        "female child",
+    ],
+    "kitchen": [
+        "kitchen",
+    ],
+    "exterior": ["exterior", "outside", "outdoor", "outdoors"],
+    "faucet": ["faucet", "drain", "tap"],
+    "water": [
+        "water",
+    ],
+    "sink": [
+        "sink",
+    ],
+    "floor": ["floor", "ground", "flooring"],
+    "plate": ["plate", "dish", "platter", "saucer"],
+    "dishes on counter": ["laying on", "placed on", "spread on"],
+    "counter": ["counter", "worktop", "counter top", "countertop"],
+    "cookies": ["cookies", "cookie", "biscuits", "biscuit"],
+    "jar": ["jar", "pot", "tin"],
+    "cabinet": ["cabinet", "cupboard", "cabinets", "cupboards"],
+    "stool": ["stool", "footstool"],
+    "window": [
+        "window",
+    ],
+    "curtain": ["curtain", "blind", "screen", "shutter"],
+    "boy_taking_cookie": [
+        "taking",
+        "grabbing",
+        "reaching",
+        "take",
+        "grab",
+        "reach",
+        "get hold",
+    ],
+    "boy_stool_falling": ["falling", "tipping", "fall", "tip"],
+    "mother_drying_washing_dishes": [
+        "dry",
+        "drying",
+        "wash",
+        "washing",
+        "clean",
+        "cleaning",
+        "scrub",
+        "scrubbing",
+        "wiping",
+        "doing dishes",
+        "doing the dishes",
+    ],
+    "water_overflowing": [
+        "overflow",
+        "overflowing",
+        "spill",
+        "spilling",
+        "cascading",
+        "sink running over",
+        "overrunning",
+    ],
+    "girl_asking_for_cookie": [
+        "ask",
+        "asking",
+        "want",
+        "wants",
+        "requesting",
+        "requests",
+    ],
+    "mother_unconcerned_about_overflowing": [
+        "unconcerned",
+        "unbothered",
+        "care",
+        "not bothered",
+        "noticed",
+        "bother",
+    ],
+    "mother_not_noticing_children": [
+        "unnoticing",
+        "noticed",
+        "not noticed",
+        "not take notice",
+        "unbothered",
+        "not bothered",
+    ],
+}
+
+# define Information Content Units (ICU) with the grid from Jensen, Chenery, Copland (2006) J Comm Dis
+ICU_picnic_en = {
+    "father": ["father", "husband", "daddy", "Dad", "dad"],
+    "mother": ["mother", "wife", "mom", "mommy", "girl"],
+    "couple": ["couple", "parents", "owners"],
+    "child": [
+        "child",
+    ],
+    "boy": ["boy", "brother"],
+    "fisherman": ["fisherman", "fisher", "rod"],
+    "sailors": ["sailors", "people", "people sailing"],
+    "dog": ["dog", "doggy", "animal"],
+    "garage": [
+        "garage",
+    ],
+    "water": [
+        "water",
+        "lake",
+        "river",
+        "sea",
+        "ocean",
+        "bank",
+        "edge",
+        "shore",
+        "riverside",
+    ],
+    "beach": ["beach", "sand"],
+    #'pail': ['pail', 'blanket'], this one is not in Jensen
+    "jetty": ["jetty", "wharf", "dock"],
+    "kite": [
+        "kite",
+    ],
+    "book": ["book", "novel"],
+    "bucket": [
+        "bucket",
+    ],
+    "spade": ["spade", "shovel"],
+    "car": ["car", "vehicle", "automobile", "sedan", "wagon"],
+    "boat": ["boat", "sailboat", "yacht", "vessel", "slinghy", "sloop"],
+    "flag": ["flag", "pole"],
+    "radio": ["radio", "transistor"],
+    "sandals": ["sandals", "shoes", "shoe", "flip", "flops"],
+    "tree": ["tree", "oak"],
+    "basket": ["basket"],
+    "drink": ["drink", "bottle", "wine", "soda", "beer", "liquor", "coffee", "tea"],
+    "house": ["house", "cottage", "shack", "cottage"],
+    "reading": ["reading", "read"],
+    "fishing": ["fishing", "fish", "catch"],
+    "pouring": ["pouring", "pour", "serve"],
+    "flying": ["flying", "fly"],  # flag #kite
+    "playing": [
+        "playing",
+        "play",
+        "build",
+        "building",
+        "blasting",
+    ],  # child on the beach # radio
+    "picnicking": ["picnic", "picnicking"],
+    "sailing": ["sailing", "sail", "boating"],
+    "parking": ["parking", "parked", "park"],
+}
+
+ICU_references = {
+    "cookie_en": ICU_cookie_en,
+    "cookie_fr": ICU_cookie_fr,
+    "picnic_en": ICU_picnic_en,
+}

speechmetryflow/semantic/icu.py

@@ -0,0 +1,33 @@
+from speechmetryflow.semantic.assets import ICU_references
+
+
+def compute_icu(text, lang, task):
+    """
+    Analyses the text to detect the presence of ICUs (Information Content Unique)
+    associated with specific subjects, places, objects, and actions.
+
+    Args:
+        text (str): The text to be analysed.
+        language (str): The language of the text ("fr" or "en").
+        task (str): Image used to generate the text ("cookie" or "picnic").
+
+    Returns:
+        dict: A dictionary containing information about the presence of ICUs. Each ICU is associated with a key
+        and the corresponding value is True if the ICU is found in the text, otherwise False.
+    """
+    ICU_dict = ICU_references.get(f"{task}_{lang}")
+
+    if ICU_dict is None:
+        return
+
+    # Initialising the results dictionary
+    icu_data = {f"icu_{icu}": False for icu in ICU_dict.keys()}
+
+    # Verification of the presence of each ICU in the text
+    for icu, words in ICU_dict.items():
+        for word in words:
+            if word in text:
+                icu_data[f"icu_{icu}"] = True
+                break  # Move to the next ICU as soon as a match is found
+
+    return icu_data

speechmetryflow/semantic/idea_density.py

@@ -0,0 +1,78 @@
+import itertools
+import numpy as np
+import torch
+
+from numpy.lib.stride_tricks import sliding_window_view
+from torch.nn.functional import cosine_similarity
+
+
+def compute_idea_density(tokens, window_shapes):
+    """
+    Compute average cosine similarities between token embeddings across
+    different sliding window sizes.
+
+    This function measures how semantically similar nearby tokens are,
+    for multiple window lengths (e.g. 3, 5, 7). For each window shape,
+    all pairwise cosine similarities are computed between embeddings in
+    that window, and the mean similarity is recorded.
+
+    To optimize performance, previously computed pairwise similarities
+    are cached in `cosine_similarity_data` so that redundant comparisons
+    across overlapping windows are not recomputed.
+
+    Parameters
+    ----------
+    tokens : list
+        A list of token objects, each exposing a `.vector` attribute
+        (NumPy array) representing its embedding.
+    window_shapes : iterable of int
+        A list or sequence of integer window sizes to apply
+        (e.g., [3, 10, 25, 40]).
+
+    Returns
+    -------
+    dict
+        A dictionary mapping each window size key
+        (e.g., "idea_density_3") to the mean cosine similarity
+        of all windows of that size. If a window shape is larger
+        than the number of tokens, the value is `None`.
+    """
+    # Convert all token vectors to PyTorch tensors
+    embeddings = [torch.from_numpy(token.vector) for token in tokens]
+    embeddings_index = list(range(len(embeddings)))
+    cosine_similarity_data = {}
+
+    results = {}
+    for window_shape in window_shapes:
+        results_key = f"idea_density_{window_shape}"
+
+        try:
+            # Generate sliding windows of token indices
+            windows = sliding_window_view(embeddings_index, window_shape=window_shape)
+        except ValueError:
+            # if window_shape is larger than embeddings
+            results[results_key] = None
+            continue
+
+        similarities = []
+        for window in windows:
+            window_cosine_similarities = []
+
+            # Compute all pairwise cosine similarities in the window
+            # cosine similarities are stored in cosine_similarity_data cache
+            # in order to avoid multiple computing
+            for col1, col2 in itertools.combinations(window, 2):
+                col_pairs = (col1, col2)
+                if col_pairs not in cosine_similarity_data:
+                    cosine_similarity_data[col_pairs] = float(
+                        cosine_similarity(embeddings[col1], embeddings[col2], dim=0)
+                    )
+                window_cosine_similarities.append(cosine_similarity_data[col_pairs])
+
+            # Average similarity for this window
+            similarities.append(np.mean(window_cosine_similarities))
+
+        # Mean similarity across all windows of this shape
+        results[results_key] = np.mean(similarities)
+
+    return results

speechmetryflow/speech_production/__init__.py

@@ -0,0 +1,62 @@
+from speechmetryflow.speech_production import fluency
+from speechmetryflow.speech_production.fragment import Fragment
+
+from collections import Counter
+
+
+def compute_counts(words):
+    """
+    Compute various count statistics for a list of words.
+
+    Parameters
+    ----------
+    words (list of str): The list of words to analyze.
+
+    Returns
+    -------
+    n (int): Total number of words in the list.
+    n_different (int): Number of distinct words (vocabulary size).
+    n_unique (int): Number of words that occur exactly once.
+    n_repetition (int): Number of distinct words that occur more than once
+
+    Examples
+    --------
+    >>> compute_counts(["cat", "dog", "cat", "mouse"])
+    (4, 3, 2, 1)
+    """
+    n = len(words)
+    counter = Counter(words)
+    n_different = len(counter)
+    n_unique = len([word for word, count in counter.items() if count == 1])
+    n_repetition = n_different - n_unique
+    return n, n_different, n_unique, n_repetition
+
+
+def metrics(raw_text, tokens, lemmas, lang, task):
+    fragment = Fragment(lang, task)
+
+    n_tokens, n_different_tokens, n_unique_tokens, n_repetition_tokens = compute_counts(
+        tokens
+    )
+    n_lemmas, n_different_lemmas, n_unique_lemmas, n_repetition_lemmas = compute_counts(
+        lemmas
+    )
+
+    return {
+        "n_tokens": n_tokens,
+        "n_different_tokens": n_different_tokens,
+        "n_unique_tokens": n_unique_tokens,
+        "n_repetition_tokens": n_repetition_tokens,
+        "n_lemmas": n_lemmas,
+        "n_different_lemmas": n_different_lemmas,
+        "n_unique_lemmas": n_unique_lemmas,
+        "n_repetition_lemmas": n_repetition_lemmas,
+        "n_fragments": fragment.counter(tokens),
+        "n_fragments_context": fragment.context(tokens),
+        "n_silent_pauses": fluency.count_silent_pauses(raw_text, lang),
+        "n_filled_pauses": fluency.count_filled_pauses(raw_text, lang),
+        "n_ucsf_silent_pauses": fluency.count_silent_pauses(
+            raw_text, pause_marker="..."
+        ),
+        "n_ucsf_filled_pauses": fluency.count_filled_pauses(raw_text, "ucsf"),
+    }

speechmetryflow/speech_production/assets/__init__.py

@@ -0,0 +1,58 @@
+# Fluency
+filled_pauses = {
+    "en": [
+        "uhm",
+        "uh",
+        "uhummm",
+        "hum",
+        "hummmm",
+        "humm",
+        "mm",
+        "mmm",
+        "Mm",
+        "um",
+        "hmmm",
+        "hmm",
+        "hm",
+        "eh",
+        "err",
+    ],
+    "fr": ["euh", "hum", "heu", "hm", "öhm", "uhm", "mmh", "mh"],
+    "ucsf": ["um", "er", "uh", "ah", "hmm", "erm", "like", "so", "well"],
+}
+
+# context fragments computation
+target_words = {
+    "en": [
+        ("the", "there"),
+        ("dry", "or"),
+        ("out", "outside"),
+        ("wash", "drying"),
+        ("curt", "the"),
+        ("is", "spill"),
+        ("wash", "is"),
+        ("bow", "Cup"),
+        ("tap", "taps"),
+        ("spill", "is"),
+        ("kit", "the"),
+        ("stand", "standing"),
+        ("go", "reaching"),
+        ("dish", "The"),
+        ("look", "uh"),
+        ("dry", "or"),
+        ("watch", "washing"),
+        ("sleeve", "sleeveless"),
+        ("cook", "cookie"),
+        ("a", "an"),
+        ("day", "daytime"),
+        ("curt", "the"),
+        ("look", "reaching"),
+        ("sister", "reach"),
+        ("cab", "the"),
+        ("cook", "the"),
+        ("stand", "climbing"),
+        ("her", "ask"),
+        ("ask", "asking"),
+        ("up", "upper"),
+    ],
+}