AI large model training data package (music category 13500G)

• American audio data package (3500G): mainly music, songs, performances, online audio, etc. from the United States, of which 90% is audio and 10% is accompanying text and video related content.

• European audio data package (2900G): mainly includes music, songs, performances, online audio, etc. from Europe, of which 85% is audio and 15% is accompanying text and video related content.

• Asian audio data package (3900G): mainly from music, songs, performances, online audio, etc. in Asia (China, Japan, South Korea, Thailand, Singapore, etc.), of which 80% is audio and 20% is accompanying text and video related content.

• Pure Chinese audio data package (2900G): mainly music and audio content from China, Taiwan, Hong Kong, Macau and other regions.

• Pure English audio data package (3500G): mainly music and audio content from the UK, Australia, etc.

• Pure Japanese audio data package (3000G): mainly music and audio content in Japan.

Data collection and compilation:

• Multi-channel collection : The data comes from 100 million audios from more than 12,000 online music platforms, music schools, audio websites, private audio collections, music institutions, etc. around the world to ensure the breadth and representativeness of the data.
• Screening by a professional team : A team of experts in the fields of musicology, literature, linguistics, etc. screens and verifies the collected data to ensure the accuracy and high quality of the data.
• Multi-level classification : Data is classified and organized according to multiple dimensions such as music type, region, era, language, etc., so that users can quickly locate the required data according to their needs.
• Data Type : This data contains data sets focused on music text adjustment, including text, pictures, audio, video and other data.

The LLM large model training data package (music category) contains the following fields:

• Track sources : from more than 1,000 music platforms around the world.
• Artist : The information of the singer or composer of the corresponding track.
• Generate music score : Music score data generated by combining music theory and arrangement instructions.
• Lyrics : Lyrics of the original data, and the edited lyrics content.
• Audio : Contains complete songs, audio of disassembled songs, etc.
• expected_metadata : The real metadata or music information provided in the original dataset.
• predict_metadata : Metadata predicted by the Mixtral model in the solution (such as tempo, tonality, etc.).
• error_message : If the code is not used, it will show <not_executed>; otherwise it will be empty or contain the exception message from the corresponding code block. The string timeout indicates that the code block execution time exceeds 10 seconds. In the current dataset version, any error or timeout will stop the generation.
• is_correct : The scoring script determines whether the final metadata is correct.
• Dataset : neuronicx1000 or OpenAI-music.
• generation_type : without_reference_solution or masked_reference_solution.

Data features:

• Diversified data sources : Covering various types of data such as pop music, classical music, jazz, electronic music, etc., to ensure the adaptability of the model in different music styles.
• High quality and low repetition rate : All data are screened by a professional team, with a repetition rate of less than 0.5%, ensuring the novelty and diversity of the training data.
• Multi-language support : mainly covers Chinese and English data, supporting the multi-language needs of global music AI projects.
• Rich audio features : Provides detailed audio analysis data, including rhythm, tonality, harmony, timbre, etc., to help the model deeply understand the music structure.
• Data privacy and compliance : Strictly comply with music copyright and data privacy regulations in various countries to ensure the legality and security of data use.

Optimization and debugging

During the model training process, adjust the model parameters, optimizer, learning rate, etc. according to the preliminary results to improve the accuracy and performance of the model. Compare the impact of different types of music data on the model effect to ensure that the required music knowledge points are fully covered and optimize the performance of the model in actual music applications.

Output and Application

After the model training is completed, it can be applied to multiple practical scenarios, such as intelligent composition systems, music recommendation platforms, lyrics generation tools, audio analysis and classification, etc. The multi-language and multi-type data in the data package supports a wide range of application needs, especially for AI projects involving the global music field. With this data package, you will easily obtain high-quality music data in multiple languages and types, helping your AI model achieve excellent performance in the music field.

When purchasing multiple data packages on the official website, you can use the following discount codes to get discounts.

• 10% discount code: LLM10 (use when purchasing 2 Chegg data packages to get a 10% discount)
• 20% discount code: LLM20 (use when purchasing 4 Chegg data packages to get a 20% discount)
• 30% discount code: LLM30 (use when purchasing 6 Chegg data packages to get a 30% discount)
• 40% discount code: LLM40 (use when purchasing 8 Chegg data packages to get a 40% discount)
• 50% discount code: LLM50 (use when purchasing 10 Chegg data packages to get a 50% discount)

Note: If the amount of the self-service order placed on the official website is large, it may not be possible to pay, and you need to contact customer service to obtain a large payment method.