alibabasglab commited on
Commit
3b7b7eb
Β·
verified Β·
1 Parent(s): 1d419c6

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +76 -3
README.md CHANGED
@@ -1,3 +1,76 @@
1
- ---
2
- license: apache-2.0
3
- ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: apache-2.0
3
+ ---
4
+ LJSpeech-1.1 High-Resolution Dataset (48,000 Hz)
5
+
6
+ Overview
7
+
8
+ The LJSpeech-1.1 dataset, widely recognized for its utility in text-to-speech (TTS) and other speech processing tasks, has now been enhanced through a cutting-edge speech super-resolution algorithm. The original dataset, which featured a sampling rate of 22,050 Hz, has been upscaled to 48,000 Hz, providing a high-fidelity version suitable for advanced audio processing tasks.
9
+
10
+ Key Features
11
+
12
+ - High-Resolution Audio: The dataset now offers audio files at a sampling rate of 48,000 Hz, delivering enhanced perceptual quality with richer high-frequency details.
13
+
14
+ - Original Content Integrity: The original linguistic content and annotation structure remain unchanged, ensuring compatibility with existing workflows.
15
+
16
+ - Broader Application Scope: Suitable for professional-grade audio synthesis, TTS systems, and other high-quality audio applications.
17
+
18
+ - Open Source: Freely available for academic and research purposes, fostering innovation in the speech and audio domains.
19
+
20
+ Enhancement Details
21
+
22
+ Original Dataset
23
+
24
+ - Source: The original LJSpeech-1.1 dataset contains 13,100 audio clips of a single female speaker reading passages from public domain books.
25
+
26
+ - Duration: Approximately 24 hours of speech data.
27
+
28
+ - Annotations: Each audio clip is paired with a corresponding text transcript.
29
+
30
+ Super-Resolution Processing
31
+
32
+ The original 22,050 Hz audio recordings were processed using a state-of-the-art MossFormer2-based speech super-resolution model. This model employs:
33
+
34
+ - Advanced Neural Architectures: A combination of transformer-based sequence modeling and convolutional networks.
35
+
36
+ - Perceptual Optimization: Loss functions designed to preserve the naturalness and clarity of speech.
37
+
38
+ - High-Frequency Reconstruction: Algorithms specifically tuned to recover lost high-frequency components, ensuring smooth and artifact-free enhancement.
39
+
40
+ Output Format
41
+
42
+ - Sampling Rate: 48,000 Hz
43
+
44
+ - Audio Format: WAV
45
+
46
+ - Bit Depth: 16-bit
47
+
48
+ - Channel Configuration: Mono
49
+
50
+ Use Cases
51
+
52
+ 1. Text-to-Speech (TTS) Synthesis
53
+ β—‹ Train high-fidelity TTS systems capable of generating human-like speech.
54
+ β—‹ Enable expressive and emotionally nuanced TTS outputs.
55
+
56
+ 2. Speech Super-Resolution Benchmarking
57
+ β—‹ Serve as a reference dataset for evaluating speech super-resolution algorithms.
58
+ β—‹ Provide a standardized benchmark for perceptual quality metrics.
59
+
60
+ 3. Audio Enhancement and Restoration
61
+ β—‹ Restore low-resolution or degraded speech signals for professional applications.
62
+ β—‹ Create high-quality voiceovers and narration for multimedia projects.
63
+
64
+ File Structure
65
+
66
+ The dataset retains the original LJSpeech-1.1 structure, ensuring ease of use:
67
+
68
+ ```sh
69
+ LJSpeech-1.1-48kHz/
70
+ β”œβ”€β”€ metadata.csv # Text transcriptions and audio file mappings
71
+ β”œβ”€β”€ wavs/ # Directory containing 48,000 Hz WAV files
72
+ └── LICENSE.txt # License information
73
+ ```
74
+ Licensing
75
+
76
+ The LJSpeech-1.1 High-Resolution Dataset is released under the same open license as the original LJSpeech-1.1 dataset (https://keithito.com/LJ-Speech-Dataset/). Users are free to use, modify, and share the dataset for academic and non-commercial purposes, provided proper attribution is given.