Redirect TIMIT download from LDC (#4145)

* redirect TIMIT download to LDC

* mention manual download in the dataset card

Commit from https://github.com/huggingface/datasets/commit/1004f364fe6e85290889c32acd2b3463e785c5a3

README.md

@@ -59,13 +59,22 @@ paperswithcode_id: timit
 
 The TIMIT corpus of read speech is designed to provide speech data for acoustic-phonetic studies and for the development and evaluation of automatic speech recognition systems. TIMIT contains broadband recordings of 630 speakers of eight major dialects of American English, each reading ten phonetically rich sentences. The TIMIT corpus includes time-aligned orthographic, phonetic and word transcriptions as well as a 16-bit, 16kHz speech waveform file for each utterance. Corpus design was a joint effort among the Massachusetts Institute of Technology (MIT), SRI International (SRI) and Texas Instruments, Inc. (TI). The speech was recorded at TI, transcribed at MIT and verified and prepared for CD-ROM production by the National Institute of Standards and Technology (NIST).
 
+The dataset needs to be downloaded manually from https://catalog.ldc.upenn.edu/LDC93S1:
+
+```
+To use TIMIT you have to download it manually.
+Please create an account and download the dataset from https://catalog.ldc.upenn.edu/LDC93S1
+Then extract all files in one folder and load the dataset with:
+`datasets.load_dataset('timit_asr', data_dir='path/to/folder/folder_name')`
+```
+
 ### Supported Tasks and Leaderboards
 
 - `automatic-speech-recognition`, `speaker-identification`: The dataset can be used to train a model for Automatic Speech Recognition (ASR). The model is presented with an audio file and asked to transcribe the audio file to written text. The most common evaluation metric is the word error rate (WER). The task has an active leaderboard which can be found at https://paperswithcode.com/sota/speech-recognition-on-timit and ranks models based on their WER.
 
 ### Languages
 
-The audio is in English. 
+The audio is in English.
 The TIMIT corpus transcriptions have been hand verified. Test and training subsets, balanced for phonetic and dialectal coverage, are specified. Tabular computer-searchable information is included as well as written documentation.
 
 ## Dataset Structure

dataset_infos.json

@@ -1 +0,0 @@
-{"clean": {"description": "The TIMIT corpus of reading speech has been developed to provide speech data for acoustic-phonetic research studies\nand for the evaluation of automatic speech recognition systems.\n\nTIMIT contains high quality recordings of 630 individuals/speakers with 8 different American English dialects,\nwith each individual reading upto 10 phonetically rich sentences.\n\nMore info on TIMIT dataset can be understood from the \"README\" which can be found here:\nhttps://catalog.ldc.upenn.edu/docs/LDC93S1/readme.txt\n", "citation": "@inproceedings{\n  title={TIMIT Acoustic-Phonetic Continuous Speech Corpus},\n  author={Garofolo, John S., et al},\n  ldc_catalog_no={LDC93S1},\n  DOI={https://doi.org/10.35111/17gk-bn40},\n  journal={Linguistic Data Consortium, Philadelphia},\n  year={1983}\n}\n", "homepage": "https://catalog.ldc.upenn.edu/LDC93S1", "license": "", "features": {"file": {"dtype": "string", "id": null, "_type": "Value"}, "audio": {"sampling_rate": 16000, "mono": true, "decode": true, "id": null, "_type": "Audio"}, "text": {"dtype": "string", "id": null, "_type": "Value"}, "phonetic_detail": {"feature": {"start": {"dtype": "int64", "id": null, "_type": "Value"}, "stop": {"dtype": "int64", "id": null, "_type": "Value"}, "utterance": {"dtype": "string", "id": null, "_type": "Value"}}, "length": -1, "id": null, "_type": "Sequence"}, "word_detail": {"feature": {"start": {"dtype": "int64", "id": null, "_type": "Value"}, "stop": {"dtype": "int64", "id": null, "_type": "Value"}, "utterance": {"dtype": "string", "id": null, "_type": "Value"}}, "length": -1, "id": null, "_type": "Sequence"}, "dialect_region": {"dtype": "string", "id": null, "_type": "Value"}, "sentence_type": {"dtype": "string", "id": null, "_type": "Value"}, "speaker_id": {"dtype": "string", "id": null, "_type": "Value"}, "id": {"dtype": "string", "id": null, "_type": "Value"}}, "post_processed": null, "supervised_keys": {"input": "file", "output": "text"}, "task_templates": [{"task": "automatic-speech-recognition", "audio_column": "audio", "transcription_column": "text"}], "builder_name": "timit_asr", "config_name": "clean", "version": {"version_str": "2.0.1", "description": "", "major": 2, "minor": 0, "patch": 1}, "splits": {"train": {"name": "train", "num_bytes": 6076580, "num_examples": 4620, "dataset_name": "timit_asr"}, "test": {"name": "test", "num_bytes": 2202968, "num_examples": 1680, "dataset_name": "timit_asr"}}, "download_checksums": {"https://data.deepai.org/timit.zip": {"num_bytes": 869007403, "checksum": "b79af42068b53045510d86854e2239a13ff77c4bd27803b40c28dce4bb5aeb0d"}}, "download_size": 869007403, "post_processing_size": null, "dataset_size": 8279548, "size_in_bytes": 877286951}}

dummy/clean/2.0.1/dummy_data.zip

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1d415bf8a373d2b6304c0e866936e0e3df530fe6ee2b0308d5965dbf4f2b4fd7
-size 292805
+oid sha256:ff277a8b6f09d34eceb60724a782412b8ff38465e8a60da2ce1eeaded644c60f
+size 294704

timit_asr.py

@@ -18,8 +18,7 @@
 
 
 import os
-
-import pandas as pd
+from pathlib import Path
 
 import datasets
 from datasets.tasks import AutomaticSpeechRecognition
@@ -47,7 +46,6 @@ More info on TIMIT dataset can be understood from the "README" which can be foun
 https://catalog.ldc.upenn.edu/docs/LDC93S1/readme.txt
 """
 
-_URL = "https://data.deepai.org/timit.zip"
 _HOMEPAGE = "https://catalog.ldc.upenn.edu/LDC93S1"
 
 
@@ -71,6 +69,15 @@ class TimitASR(datasets.GeneratorBasedBuilder):
 
     BUILDER_CONFIGS = [TimitASRConfig(name="clean", description="'Clean' speech.")]
 
+    @property
+    def manual_download_instructions(self):
+        return (
+            "To use TIMIT you have to download it manually. "
+            "Please create an account and download the dataset from https://catalog.ldc.upenn.edu/LDC93S1 \n"
+            "Then extract all files in one folder and load the dataset with: "
+            "`datasets.load_dataset('timit_asr', data_dir='path/to/folder/folder_name')`"
+        )
+
     def _info(self):
         return datasets.DatasetInfo(
             description=_DESCRIPTION,
@@ -106,42 +113,30 @@ class TimitASR(datasets.GeneratorBasedBuilder):
         )
 
     def _split_generators(self, dl_manager):
-        archive_path = dl_manager.download_and_extract(_URL)
 
-        train_csv_path = os.path.join(archive_path, "train_data.csv")
-        test_csv_path = os.path.join(archive_path, "test_data.csv")
+        data_dir = os.path.abspath(os.path.expanduser(dl_manager.manual_dir))
+
+        if not os.path.exists(data_dir):
+            raise FileNotFoundError(
+                f"{data_dir} does not exist. Make sure you insert a manual dir via `datasets.load_dataset('timit_asr', data_dir=...)` that includes files unzipped from the TIMIT zip. Manual download instructions: {self.manual_download_instructions}"
+            )
 
         return [
-            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"data_info_csv": train_csv_path}),
-            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"data_info_csv": test_csv_path}),
+            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"split": "train", "data_dir": data_dir}),
+            datasets.SplitGenerator(name=datasets.Split.TEST, gen_kwargs={"split": "test", "data_dir": data_dir}),
         ]
 
-    def _generate_examples(self, data_info_csv):
+    def _generate_examples(self, split, data_dir):
         """Generate examples from TIMIT archive_path based on the test/train csv information."""
-        # Extract the archive path
-        data_path = os.path.join(os.path.dirname(data_info_csv).strip(), "data")
-
-        # Read the data info to extract rows mentioning about non-converted audio only
-        data_info = pd.read_csv(open(data_info_csv, encoding="utf8"))
-        # making sure that the columns having no information about the file paths are removed
-        data_info.dropna(subset=["path_from_data_dir"], inplace=True)
-
-        # filter out only the required information for data preparation
-        data_info = data_info.loc[(data_info["is_audio"]) & (~data_info["is_converted_audio"])]
-
         # Iterating the contents of the data to extract the relevant information
-        for audio_idx in range(data_info.shape[0]):
-            audio_data = data_info.iloc[audio_idx]
-
-            # extract the path to audio
-            wav_path = os.path.join(data_path, *(audio_data["path_from_data_dir"].split("/")))
+        for wav_path in sorted(Path(data_dir).glob(f"**/{split.upper()}/**/*.WAV")):
 
             # extract transcript
-            with open(wav_path.replace(".WAV", ".TXT"), encoding="utf-8") as op:
+            with open(wav_path.with_suffix(".TXT"), encoding="utf-8") as op:
                 transcript = " ".join(op.readlines()[0].split()[2:])  # first two items are sample number
 
             # extract phonemes
-            with open(wav_path.replace(".WAV", ".PHN"), encoding="utf-8") as op:
+            with open(wav_path.with_suffix(".PHN"), encoding="utf-8") as op:
                 phonemes = [
                     {
                         "start": i.split(" ")[0],
@@ -152,7 +147,7 @@ class TimitASR(datasets.GeneratorBasedBuilder):
                 ]
 
             # extract words
-            with open(wav_path.replace(".WAV", ".WRD"), encoding="utf-8") as op:
+            with open(wav_path.with_suffix(".WRD"), encoding="utf-8") as op:
                 words = [
                     {
                         "start": i.split(" ")[0],
@@ -162,16 +157,21 @@ class TimitASR(datasets.GeneratorBasedBuilder):
                     for i in op.readlines()
                 ]
 
+            dialect_region = wav_path.parents[1].name
+            sentence_type = wav_path.name[0:2]
+            speaker_id = wav_path.parents[0].name[1:]
+            id_ = wav_path.stem
+
             example = {
-                "file": wav_path,
-                "audio": wav_path,
+                "file": str(wav_path),
+                "audio": str(wav_path),
                 "text": transcript,
                 "phonetic_detail": phonemes,
                 "word_detail": words,
-                "dialect_region": audio_data["dialect_region"],
-                "sentence_type": audio_data["filename"][0:2],
-                "speaker_id": audio_data["speaker_id"],
-                "id": audio_data["filename"].replace(".WAV", ""),
+                "dialect_region": dialect_region,
+                "sentence_type": sentence_type,
+                "speaker_id": speaker_id,
+                "id": id_,
             }
 
-            yield audio_idx, example
+            yield id_, example