ser-standardizer 0.1.0__tar.gz

ser_standardizer-0.1.0/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2025 Vinicius Suaiden
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

ser_standardizer-0.1.0/PKG-INFO

@@ -0,0 +1,166 @@
+Metadata-Version: 2.4
+Name: ser-standardizer
+Version: 0.1.0
+Summary: A tool for standardizing Speech Emotion Recognition (SER) datasets and extracting acoustic features.
+Author-email: Vinicius Suaiden <vinicius.suaiden@usp.br>, Miguel Arjona Ramirez <maramire@usp.br>, Wesley Beccaro <wesleybeccaro@usp.br>
+License: MIT License
+        
+        Copyright (c) 2025 Vinicius Suaiden
+        
+        Permission is hereby granted, free of charge, to any person obtaining a copy
+        of this software and associated documentation files (the "Software"), to deal
+        in the Software without restriction, including without limitation the rights
+        to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+        copies of the Software, and to permit persons to whom the Software is
+        furnished to do so, subject to the following conditions:
+        
+        The above copyright notice and this permission notice shall be included in all
+        copies or substantial portions of the Software.
+        
+        THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+        IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+        FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+        AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+        LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+        OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+        SOFTWARE.
+        
+Project-URL: Homepage, https://github.com/ViniciusSuaiden/ser-standardizer
+Project-URL: Repository, https://github.com/ViniciusSuaiden/ser-standardizer
+Project-URL: Issues, https://github.com/ViniciusSuaiden/ser-standardizer/issues
+Classifier: Programming Language :: Python :: 3
+Classifier: License :: OSI Approved :: MIT License
+Classifier: Operating System :: OS Independent
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.8
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: pandas>=1.3.0
+Requires-Dist: audb>=1.0.0
+Provides-Extra: features
+Requires-Dist: numpy<3.0,>=1.21.0; extra == "features"
+Requires-Dist: soundfile>=0.12.0; extra == "features"
+Requires-Dist: scipy>=1.10.0; extra == "features"
+Requires-Dist: ipython>=9.0.0; extra == "features"
+Requires-Dist: ipywidgets>=8.1.8; extra == "features"
+Requires-Dist: noisereduce>=3.0.3; extra == "features"
+Requires-Dist: tqdm>=4.60.0; extra == "features"
+Requires-Dist: opensmile>=2.4.0; extra == "features"
+Dynamic: license-file
+
+# SER-Standardizer
+
+🇺🇸 English | [🇧🇷 Português](README.pt-br.md)
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+[![Python Version](https://img.shields.io/badge/python-3.8%2B-blue)](https://www.python.org/downloads/)
+
+**A modular infrastructure for dataset standardization and feature extraction in Speech Emotion Recognition (SER).**
+
+**SER-Standardizer** is a Python package built to unify the formatting of multiple speech databases, mitigating metadata heterogeneity. Its main goal is to facilitate *cross-corpus training* and generalization testing of machine learning models.
+
+Beyond the standardization pipeline, the tool provides an extension module for acoustic feature extraction (spectral and prosodic) by integrating the **openSMILE** framework.
+
+## 📋 Currently Supported Datasets
+
+The tool currently supports loading and standardizing the following databases:
+
+* **CREMA-D** | **IEMOCAP** | **EmoUERJ** (PT-BR) | **MSP-IMPROV** | **MSP-PODCAST** | **RAVDESS** | **EmoDB** (DE)
+
+Per-dataset download and directory-layout guides are available under [`docs/en/`](docs/en/).
+
+## 🚀 Installation
+
+### Prerequisites
+* Python 3.8 or higher
+* Libraries listed in `pyproject.toml` (installed automatically).
+
+### Installation from source
+Clone this repository to install it with `pip` (read more about it below):
+```
+git clone https://github.com/ViniciusSuaiden/ser-standardizer.git
+cd ser-standardizer
+```
+
+The library was designed with flexibility in mind, offering two installation modes to optimize the use of computational resources:
+
+#### 1. Base Installation (Lightweight Metadata)
+Installs only the essential packages (`pandas`, `audb`) for tabular standardization and filtering.
+```bash
+pip install .
+```
+#### 2. Full Installation (With Signal Processing)
+Installs the heavy dependencies required for the feature extraction and audio manipulation module (opensmile, noisereduce).
+```bash
+pip install '.[features]'
+```
+### 💻 How to Use
+The workflow is split into three logical steps:
+
+**1. Preprocessing (CLI)**
+
+After installation, the `ser-std` command becomes available in your terminal.
+To standardize a specific dataset:
+```bash
+# Example: crema_d
+ser-std --dataset crema_d --input_dir /path/to/crema
+```
+The standardized `.csv` file is written to the user's home folder, with a specific name for each database.
+
+**2. Data Handling and Filtering (Python)**
+
+After preprocessing, use the library to load, filter and manipulate the audio directly in your code or Jupyter Notebook.
+```python
+import ser_standardizer as ser
+
+# Loads multiple standardized datasets into a single DataFrame
+df = ser.load_datasets(["crema_d", "ravdess", "iemocap"])
+
+# Filters by dataset, emotion and gender
+df_target = ser.filters(
+    df,
+    datasets=['ravdess', 'iemocap'],
+    emotions=['anger', 'sad'],
+    genders=['female']
+)
+```
+**3. Acoustic Feature Extraction (Requires [features])**
+
+The extraction module automates digital signal processing. It includes the option to apply a Voice Activity Detection (VAD) mask based on RMS energy to extract features strictly from the effective phonation segments.
+```python
+# Batch extraction via openSMILE
+# Supported 'feature_set' values: 'eGeMAPS' or 'ComParE'
+features_df = ser.extract_features(
+    df_target,
+    feature_set='eGeMAPS',
+    feature_level='functionals', # 'functionals' (1 vector per utterance) or 'llds' (per frame)
+    use_vad=False, # VAD (silence removal) is only allowed with feature_level='llds'
+    sr=16000       # Sampling rate
+)
+
+# The result preserves the original indices, making concatenation easy
+import pandas as pd
+final_dataset = pd.concat([df_target, features_df], axis=1)
+
+# Group-wise normalization (z-score) — removes corpus/speaker offsets.
+# Pass the grouping column (e.g., 'dataset' or 'speaker_id').
+X_corpus  = ser.normalize(features_df, df_target['dataset'])
+X_speaker = ser.normalize(features_df, df_target['speaker_id'])
+
+# --- Extra Utilities ---
+# Listen to the audio in a Jupyter Notebook
+ser.listen(df_target, index=42)
+
+# Load the raw audio as a NumPy array (e.g., to feed neural networks)
+audio_array = ser.load_audio(df_target, index=42)
+
+# Load a batch of audio with zero-padding
+batch = ser.load_batch(df_target, begin=0, end=32)
+```
+
+### ✍️ Authors
+Vinicius Suaiden - USP - vinicius.suaiden@usp.br
+
+Miguel Arjona Ramirez - USP - maramire@usp.br
+
+Wesley Beccaro - USP - wesleybeccaro@usp.br

ser_standardizer-0.1.0/README.md

@@ -0,0 +1,116 @@
+# SER-Standardizer
+
+🇺🇸 English | [🇧🇷 Português](README.pt-br.md)
+
+[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
+[![Python Version](https://img.shields.io/badge/python-3.8%2B-blue)](https://www.python.org/downloads/)
+
+**A modular infrastructure for dataset standardization and feature extraction in Speech Emotion Recognition (SER).**
+
+**SER-Standardizer** is a Python package built to unify the formatting of multiple speech databases, mitigating metadata heterogeneity. Its main goal is to facilitate *cross-corpus training* and generalization testing of machine learning models.
+
+Beyond the standardization pipeline, the tool provides an extension module for acoustic feature extraction (spectral and prosodic) by integrating the **openSMILE** framework.
+
+## 📋 Currently Supported Datasets
+
+The tool currently supports loading and standardizing the following databases:
+
+* **CREMA-D** | **IEMOCAP** | **EmoUERJ** (PT-BR) | **MSP-IMPROV** | **MSP-PODCAST** | **RAVDESS** | **EmoDB** (DE)
+
+Per-dataset download and directory-layout guides are available under [`docs/en/`](docs/en/).
+
+## 🚀 Installation
+
+### Prerequisites
+* Python 3.8 or higher
+* Libraries listed in `pyproject.toml` (installed automatically).
+
+### Installation from source
+Clone this repository to install it with `pip` (read more about it below):
+```
+git clone https://github.com/ViniciusSuaiden/ser-standardizer.git
+cd ser-standardizer
+```
+
+The library was designed with flexibility in mind, offering two installation modes to optimize the use of computational resources:
+
+#### 1. Base Installation (Lightweight Metadata)
+Installs only the essential packages (`pandas`, `audb`) for tabular standardization and filtering.
+```bash
+pip install .
+```
+#### 2. Full Installation (With Signal Processing)
+Installs the heavy dependencies required for the feature extraction and audio manipulation module (opensmile, noisereduce).
+```bash
+pip install '.[features]'
+```
+### 💻 How to Use
+The workflow is split into three logical steps:
+
+**1. Preprocessing (CLI)**
+
+After installation, the `ser-std` command becomes available in your terminal.
+To standardize a specific dataset:
+```bash
+# Example: crema_d
+ser-std --dataset crema_d --input_dir /path/to/crema
+```
+The standardized `.csv` file is written to the user's home folder, with a specific name for each database.
+
+**2. Data Handling and Filtering (Python)**
+
+After preprocessing, use the library to load, filter and manipulate the audio directly in your code or Jupyter Notebook.
+```python
+import ser_standardizer as ser
+
+# Loads multiple standardized datasets into a single DataFrame
+df = ser.load_datasets(["crema_d", "ravdess", "iemocap"])
+
+# Filters by dataset, emotion and gender
+df_target = ser.filters(
+    df,
+    datasets=['ravdess', 'iemocap'],
+    emotions=['anger', 'sad'],
+    genders=['female']
+)
+```
+**3. Acoustic Feature Extraction (Requires [features])**
+
+The extraction module automates digital signal processing. It includes the option to apply a Voice Activity Detection (VAD) mask based on RMS energy to extract features strictly from the effective phonation segments.
+```python
+# Batch extraction via openSMILE
+# Supported 'feature_set' values: 'eGeMAPS' or 'ComParE'
+features_df = ser.extract_features(
+    df_target,
+    feature_set='eGeMAPS',
+    feature_level='functionals', # 'functionals' (1 vector per utterance) or 'llds' (per frame)
+    use_vad=False, # VAD (silence removal) is only allowed with feature_level='llds'
+    sr=16000       # Sampling rate
+)
+
+# The result preserves the original indices, making concatenation easy
+import pandas as pd
+final_dataset = pd.concat([df_target, features_df], axis=1)
+
+# Group-wise normalization (z-score) — removes corpus/speaker offsets.
+# Pass the grouping column (e.g., 'dataset' or 'speaker_id').
+X_corpus  = ser.normalize(features_df, df_target['dataset'])
+X_speaker = ser.normalize(features_df, df_target['speaker_id'])
+
+# --- Extra Utilities ---
+# Listen to the audio in a Jupyter Notebook
+ser.listen(df_target, index=42)
+
+# Load the raw audio as a NumPy array (e.g., to feed neural networks)
+audio_array = ser.load_audio(df_target, index=42)
+
+# Load a batch of audio with zero-padding
+batch = ser.load_batch(df_target, begin=0, end=32)
+```
+
+### ✍️ Authors
+Vinicius Suaiden - USP - vinicius.suaiden@usp.br
+
+Miguel Arjona Ramirez - USP - maramire@usp.br
+
+Wesley Beccaro - USP - wesleybeccaro@usp.br

ser_standardizer-0.1.0/pyproject.toml

@@ -0,0 +1,51 @@
+[build-system]
+requires = ["setuptools>=61.0"]
+build-backend = "setuptools.build_meta"
+
+[project]
+name = "ser-standardizer"
+version = "0.1.0"
+description = "A tool for standardizing Speech Emotion Recognition (SER) datasets and extracting acoustic features."
+readme = "README.md"
+authors = [
+  { name = "Vinicius Suaiden", email = "vinicius.suaiden@usp.br" },
+  { name = "Miguel Arjona Ramirez", email = "maramire@usp.br" },
+  { name = "Wesley Beccaro", email = "wesleybeccaro@usp.br" },
+]
+license = { file = "LICENSE" }
+classifiers = [
+    "Programming Language :: Python :: 3",
+    "License :: OSI Approved :: MIT License",
+    "Operating System :: OS Independent",
+    "Topic :: Scientific/Engineering :: Artificial Intelligence",
+]
+dependencies = [
+    "pandas>=1.3.0",
+    "audb>=1.0.0"
+]
+requires-python = ">=3.8"
+
+[project.optional-dependencies]
+features = [
+    "numpy>=1.21.0,<3.0",
+    "soundfile>=0.12.0",     
+    "scipy>=1.10.0",         
+    "ipython>=9.0.0",
+    "ipywidgets>=8.1.8",
+    "noisereduce>=3.0.3",
+    "tqdm>=4.60.0",
+    "opensmile>=2.4.0"
+]
+
+[project.urls]
+Homepage = "https://github.com/ViniciusSuaiden/ser-standardizer"
+Repository = "https://github.com/ViniciusSuaiden/ser-standardizer"
+Issues = "https://github.com/ViniciusSuaiden/ser-standardizer/issues"
+
+[project.scripts]
+ser-std = "ser_standardizer.main:main"
+
+[tool.setuptools.packages.find]
+where = ["src"] 
+include = ["ser_standardizer*"]
+namespaces = false

ser_standardizer-0.1.0/setup.cfg

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

ser_standardizer-0.1.0/src/ser_standardizer/__init__.py

@@ -0,0 +1 @@
+from .api import load_datasets, load_audio, load_batch, listen, filters, normalize, extract_features

ser_standardizer-0.1.0/src/ser_standardizer/api.py

@@ -0,0 +1,83 @@
+import pandas as pd
+import os
+from pathlib import Path
+
+def load_datasets(datasets):
+    if isinstance(datasets, str):
+        target_datasets = [datasets]
+    else:
+        target_datasets = datasets
+
+    dataframes = []
+    DATA_DIR = os.path.join(Path.home(), ".ser_standardizer_data")
+    
+    for name in target_datasets:
+        file_path = os.path.join(DATA_DIR, f"process_{name}.csv")
+        if not file_path or not os.path.exists(file_path):
+            raise FileNotFoundError(
+                f"O dataset '{name}' não foi encontrado. "
+                f"Certifique-se de ter rodado: ser-std --dataset {name} ..."
+            )
+        df = pd.read_csv(file_path)
+        dataframes.append(df)
+
+    if not dataframes:
+        raise ValueError("Nenhum dataset válido foi fornecido.")
+        
+    return pd.concat(dataframes, ignore_index=True)
+
+def filters(df, datasets=None, emotions=None, genders=None, languages=None, speakers=None):
+    mask = pd.Series([True] * len(df), index=df.index)
+
+    def to_list(val):
+        if isinstance(val, str):
+            return [val]
+        return val
+
+    if datasets:
+        target = [x.lower() for x in to_list(datasets)]
+        mask = mask & df['dataset'].str.lower().isin(target)
+
+    if emotions:
+        target = [x.lower() for x in to_list(emotions)]
+        mask = mask & df['emotion'].str.lower().isin(target)
+
+    if genders:
+        target = [x.lower() for x in to_list(genders)]
+        mask = mask & df['gender'].str.lower().isin(target)
+
+    if languages:
+        target = [x.lower() for x in to_list(languages)]
+        mask = mask & df['language'].str.lower().isin(target)
+
+    if speakers:
+        target = [x.lower() for x in to_list(speakers)]
+        mask = mask & df['speaker_id'].str.lower().isin(target)
+
+    print(f"Filtro aplicado. Restaram {len(df[mask])} amostras.")
+    return df[mask].copy()
+
+def normalize(features, group, eps=1e-8):
+    """Aplica z-score por grupo em `features` (ex.: por dataset ou locutor)."""
+    g = group.reindex(features.index)
+    out = features.groupby(g).transform(
+        lambda c: (c - c.mean()) / (c.std(ddof=0) + eps)
+    )
+    return out.fillna(0.0)
+
+try:
+    from .features.preprocessing import listen, load_audio, load_batch
+    from .features.extractor import extract_features
+    FEATURES_AVAILABLE = True
+
+except ImportError:
+    FEATURES_AVAILABLE = False
+    
+    def _missing_deps_error(*args, **kwargs):
+        raise ImportError(
+            "Esta função requer o submódulo de extração acústica.\n"
+            "Para utilizá-la, reinstale o pacote com as dependências completas:\n\n"
+            "    pip install '.[features]'"
+        )
+        
+    listen = load_audio = load_batch = extract_features = _missing_deps_error

ser_standardizer-0.1.0/src/ser_standardizer/features/__init__.py

@@ -0,0 +1,3 @@
+# Módulo opcional (instale com `pip install "ser-standardizer[features]"`).
+# Sem imports aqui: os submódulos são importados sob demanda em api.py,
+# dentro de try/except ImportError.

ser_standardizer-0.1.0/src/ser_standardizer/features/extractor.py

@@ -0,0 +1,97 @@
+import pandas as pd
+import numpy as np
+from tqdm.auto import tqdm
+import opensmile
+from .preprocessing import load_audio, remove_silence
+
+def extract_features(df, feature_set='eGeMAPS', feature_level='functionals', use_vad=False, sr=16000):
+    """
+    Extrai características acústicas utilizando o openSMILE.
+    
+    Args:
+        df: DataFrame contendo a coluna 'file_path'.
+        feature_set: 'eGeMAPS' ou 'ComParE' (padrões validados na literatura).
+        feature_level: 'functionals' (estatísticas globais por arquivo) ou 'llds' (nível de frame).
+        use_vad: Se True, remove silêncios baseados na energia RMS. 
+                 ATENÇÃO: Permitido apenas quando feature_level='llds'.
+        sr: Taxa de amostragem esperada.
+        
+    Returns:
+        Um novo pd.DataFrame contendo as features. 
+        Para 'functionals', mantém o índice original.
+        Para 'llds', adiciona o índice original ('file_idx') para cruzamento de metadados.
+    """
+
+    if use_vad and feature_level.lower() == 'functionals':
+        raise ValueError(
+            "Erro de Metodologia: A aplicação de VAD no áudio bruto antes da extração "
+            "de funcionais corrompe as métricas temporais padronizadas do openSMILE "
+            "(como taxa de fala e proporção de pausas) e introduz artefatos no espectro. \n"
+            "-> Para extrair 'functionals' padronizados, defina use_vad=False. \n"
+            "-> O uso de use_vad=True é permitido apenas para extração no nível de frame ('llds')."
+        )
+
+    if feature_set.lower() == 'egemaps':
+        fset = opensmile.FeatureSet.eGeMAPSv02
+    elif feature_set.lower() == 'compare':
+        fset = opensmile.FeatureSet.ComParE_2016
+    else:
+        raise ValueError("Conjunto não suportado. Escolha 'eGeMAPS' ou 'ComParE'.")
+
+    # 2. Definir o Nível da Extração
+    if feature_level.lower() == 'functionals':
+        flevel = opensmile.FeatureLevel.Functionals
+        level_name = "Functionals"
+    elif feature_level.lower() in ['lld', 'llds']:
+        flevel = opensmile.FeatureLevel.LowLevelDescriptors
+        level_name = "LowLevelDescriptors"
+    else:
+        raise ValueError("Nível não suportado. Escolha 'functionals' ou 'llds'.")
+
+    print(f"Inicializando openSMILE ({feature_set} - {level_name})...")
+    smile = opensmile.Smile(
+        feature_set=fset,
+        feature_level=flevel,
+    )
+    
+    features_list = []
+    
+    for idx in tqdm(df.index, desc=f"Extraindo características ({level_name})"):
+        file_path = df.loc[idx, 'file_path']
+        
+        try:
+            if use_vad:
+                audio = load_audio(df, idx, sr=sr)
+                audio_clean = remove_silence(audio, sr=sr)
+                
+                if len(audio_clean) == 0:
+                    audio_clean = np.zeros(sr)
+                    
+                feat_df = smile.process_signal(audio_clean, sr)
+            else:
+                feat_df = smile.process_file(file_path)
+            
+            if flevel == opensmile.FeatureLevel.Functionals:
+                feat_series = feat_df.iloc[0].rename(idx)
+                features_list.append(feat_series)
+            else:
+                feat_df['file_idx'] = idx
+                features_list.append(feat_df)
+                
+        except Exception as e:
+            print(f"Erro na extração do índice {idx}: {e}")
+            if flevel == opensmile.FeatureLevel.Functionals:
+                features_list.append(pd.Series(dtype=float, name=idx))
+            else:
+                features_list.append(pd.DataFrame({'file_idx': [idx]}))
+
+    print("Consolidando matriz de features...")
+    
+    if flevel == opensmile.FeatureLevel.Functionals:
+        result_df = pd.DataFrame(features_list)
+    else:
+        result_df = pd.concat(features_list)
+        if 'file_idx' in result_df.columns:
+            result_df = result_df.set_index('file_idx', append=True)
+            
+    return result_df

ser_standardizer-0.1.0/src/ser_standardizer/features/preprocessing.py

@@ -0,0 +1,92 @@
+import soundfile as sf
+import scipy.signal
+import numpy as np
+from IPython.display import Audio, display
+from tqdm.auto import tqdm
+import noisereduce as nr
+
+def scipy_load(path, target_sr=None):
+
+    audio, native_sr = sf.read(path)
+
+    if audio.ndim > 1:
+        audio = np.mean(audio, axis=1)
+
+    if target_sr is not None and native_sr != target_sr:
+        num_samples = round(len(audio) * float(target_sr) / native_sr)
+        audio = scipy.signal.resample(audio, num_samples)
+        return audio, target_sr
+
+    return audio, native_sr
+
+def remove_silence(audio, sr=16000, frame_length=2048, hop_length=512, threshold=0.25):
+    audio_clean = nr.reduce_noise(y=audio, sr=sr)
+
+    n_frames = 1 + (len(audio_clean) - frame_length) // hop_length
+    if n_frames <= 0:
+        return audio_clean
+
+    rms = np.array([
+        np.sqrt(np.mean(audio_clean[i*hop_length : i*hop_length + frame_length] ** 2))
+        for i in range(n_frames)
+    ])
+
+    max_rms = np.max(rms)
+    if max_rms == 0:
+        return np.array([])
+
+    rms_norm = rms / max_rms
+    speech_mask = rms_norm > threshold
+
+    sample_mask = np.zeros(len(audio_clean), dtype=bool)
+    for i, active in enumerate(speech_mask):
+        if active:
+            start = i * hop_length
+            end = min(start + hop_length, len(audio_clean))
+            sample_mask[start:end] = True
+
+    return audio_clean[sample_mask]
+
+def listen(df, index):
+    if index not in df.index:
+        print(f"Aviso: Não há nenhum áudio disponível com o índice {index}.")
+        return
+
+    path = df.loc[index, 'file_path']
+    emotion = df.loc[index, 'emotion']
+    
+    print(f"Tocando áudio {index} | Emoção: {emotion}")
+    display(Audio(filename=path))
+
+def load_audio(df, index, sr=16000):
+    if index not in df.index:
+        raise IndexError("Índice não encontrado no Dataset.")
+    path = df.loc[index, 'file_path']
+    audio, _ = scipy_load(path, target_sr=sr)
+    return audio
+
+def load_batch(df, begin, end, sr=16000, max_1000=True):
+    qtd = end - begin
+    if qtd > 1000 and max_1000:
+        raise ValueError(f"Você tentou carregar {qtd} áudios. O limite é 1000. Defina max_1000=False se tiver certeza.")
+
+    batch = []
+    subset_indices = df.index[begin:end]
+
+    for idx in tqdm(subset_indices, desc="Carregando áudios"):
+        try:
+            audio = load_audio(df, idx, sr)
+            batch.append(audio)
+        except Exception as e:
+            print(f"Erro ao carregar índice {idx}: {e}")
+            batch.append(None)
+
+    max_len = max(len(x) for x in batch if x is not None)
+    padded_batch = np.zeros((len(batch), max_len), dtype=np.float32)
+
+    for i, audio in enumerate(batch):
+        if audio is not None:
+            padded_batch[i, :audio.shape[0]] = audio
+            
+    return padded_batch
+

ser_standardizer-0.1.0/src/ser_standardizer/loaders/__init__.py

@@ -0,0 +1,21 @@
+from .crema_d import process as process_crema
+from .iemocap import process as process_iemocap
+from .msp_improv import process as process_msp
+from .msp_podcast import process as process_msp_podcast
+from .emouerj import process as process_emouerj
+from .ravdess import process as process_ravdess
+from .emodb import process as process_emodb
+
+# Dicionário que mapeia o nome do dataset (string) para a função de processamento
+DATASET_LOADERS = {
+    'crema_d': process_crema,
+    'iemocap': process_iemocap,
+    'msp_improv': process_msp,
+    'msp_podcast': process_msp_podcast,
+    'emouerj': process_emouerj,
+    'ravdess': process_ravdess,
+    'emodb': process_emodb
+}
+
+def get_available_datasets():
+    return list(DATASET_LOADERS.keys())