YouTube Audio Sample Generator - Refine SubAgent & Output

Describe

Describe your task

Refine

Refine the plan

SubAgents

Review all agents

Deploy

Deploy your agent

OVERVIEW OF AGENTS:
Purpose: Download and process YouTube audio via a series of specialized agents that handle different aspects of audio processing and sample pack creation. Each agent performs specific tasks using a combination of existing and required skills.

SUBAGENT 1: MEDIAAGENT
- Purpose: Download and process YouTube audio
- Final Output: [raw-wav-audio]
- Skills:
  - #193 Download Entire YouTube Video As MP3
  - #178 Convert 1-20 MP3s to wav
- Sequence: Download YouTube video > Convert to MP3 > Convert to WAV

SUBAGENT 2: ANALYSISAGENT
- Purpose: Detect tempo and extract stems
- Final Output: [analyzed-stem-sections]
- Skills:
  - #180 Extract Beatpoints & Tempo of MP3
  - #197 Extract Instrumental & Vocal From MP3
  - #178 Convert MP3 stems to WAV
- Sequence: Analyze tempo > Extract stems > Convert to WAV format

SUBAGENT 3: SAMPLEAGENT
- Purpose: Generate and analyze beat-aligned samples
- Final Output: [ranked-beat-samples]
- Skills:
  - #219 Cut Wav/mp3 Audio into Multiple Pieces/Samples
  - Sample Analysis Function (using librosa)
    - Analyze samples (amplitude, frequency content)
    - Rank samples by musical interest
    - Select top ~48 samples
- Sequence: Cut audio > Analyze samples > Rank samples > Select top samples

SUBAGENT 4: PACKAGEAGENT
- Purpose: Organize and deliver final samples
- Final Output: [final-sample-pack]
- Skills:
  - File Organization Function
    - Create timestamp/stem folders
    - Normalize audio levels
    - Generate metadata.json
    - Create zip archive
- Sequence: Create folders > Normalize audio > Generate metadata > Create archive

Final Exact Output You Want ⚠️IMPORTANT⚠️

Expand

Other Possible SubAgents (Consider Adding These Above) ☝️🤔

Expand

#179 - Create Visual Waveform Of 60 second Wav/mp3 File
DESCRIPTION: This text will create a visual "waveform" of an input wav file. This is useful for analyzing the "loudness" or amplitude of an audio at different points (we can pass this image to another skill such as analyzing an image with a vision model to "see" the changes of loudness throughout the duration of the audio).
INPUT NEEDED: URL of a .wav file
OUTPUT: A visual "spectrograph" waveform of wav file (as jpeg URL)

#188 - Extract 10 audio stems from mp3
DESCRIPTION: This skill will input an MP3 URL and extract all instruments and variations, then return between 1-10x MP3s. MP3s will be returned in order of volume/amplitude (ie most loudest first), and only stems with a minimum level of loudness will be returned (so if an MP3 has no woodwind instruments, then that audio will not be included). Stems that can be extracted include: Vocals, instrumental, drums, bass guitar, electric guitar, acoustic guitar, piano, synthesizer, strings, wind.
INPUT NEEDED: URL of MP3
OUTPUT: List of all 10 stems, in MP3 URL, listed in order of loudness.

#194 - Cut Small Section From MP4 Video
DESCRIPTION: This skill extracts a section from a given MP4 video.
INPUT NEEDED: Requires URL of MP4 to extract, along with timing (ideally expessed as start time to extract and duration to extract in seconds)
OUTPUT: URL of MP4 snippet

#196 - Extract MP3 Audio From MP4 File
DESCRIPTION: This skill will input an MP4 file and extract its audio as an MP3 file.
INPUT NEEDED: MP3 URL
OUTPUT: MP4 URL

#176 - Analyze An Image With GPT Vision & Return Text
DESCRIPTION: This skill takes an image, and a given instruction, and uses GPT Vision to analyze and return text about the image.
INPUT NEEDED: Image URL (plus ideally text prompt describing how image should be analysed
OUTPUT: Text response

#177 - Analyze 4x Images With GPTVision And Return 4x Text
DESCRIPTION: This task uses GPT4O "Vision" to make quick assessments of between 2-18 images at once. This task is excellent for making `quick-and-dirty` analysis of different images, then returning a brief summary of the images
INPUT NEEDED: 2-18 image URLs and the prompt (explaining the purpose of the analysis)
OUTPUT: A summary of the images, like 1. https://image1.. commentshere 2. https://image2.. etc

#215 - Summarise and Return All Final Output Files
DESCRIPTION: This skill returns a summary of various previous skills that have completed, and is particularly useful when appearing at the end of an entire Agent task, silo or sequence of skills.
INPUT NEEDED: at least one output (perhaps other supplementary ones that may be helpful)
OUTPUT: nice text summary containing 1 or more outputs

Task Summary (You Wrote)

Expand

I want to input a youtube video, then get a collection of music samples returned to me (.wav files in a zip). I think I should specify 3 specific times (eg 5 seconds, 130 seconds, 190 seconds, and then the agent  rip the audio file from YouTube, then extract stems (instrumentals, vocals), and then extract (say) 16 beats worth of audio - beginning from the 3 timings I've entered. This gives us 3x separate audio sections to sample, and if we then separate them into vocals and instrumentals that then gives us 6x separate audio files to sample from (although we will use the initial tempo/beatpoint for all 6). Next we will want to use the beatpoint and tempo to extract samples (eg 1/2/4/8 beat in lengths, beginning from the beatpoints). So we chopup the samples eg using SOX, perhaps chopping based on beat length (if BPM = 90bpm, then 1 beat is 0.66secs) and beat position, and chopping maybe a mix of 1/2/4/8 beat samples.  This will likely give us far too many samples, so we may want to finally filter them down (to "the best 48 samples"). We could do this by using a library like Librosa to filter somehow (eg based on average amplitude), and rank the samples. We will likely need to calculate how many samples we are going to get, how/if to rank them, and then generate them acccordingly. I am not sure whether we should filter or just do a bulk import. Finally we will want to save all the files to a server and zip them up

Diagram Of Entire Task (All SubAgents)

Expand Diagram

Ready To Proceed?

When you are ready to proceed, click the Save button to load the next step...
The next step is SubAgents. It will take around 10 minutes to load this page...
If you are entirely unhappy with this page, click Back to rerun Step 1 again.
Before saving, check the SubAgents and Output fields above (very important!)
The Agent heavily relies on the SubAgents and Output fields to build your task.

-- The LLM has raised some questions to help you clarify the desired final `output` (important - this must be specific) 
-- These ideas are worth scanning briefly (then either amend output/subagents -- or go back to step 1)
-- The first few points regarding how to clarify the `output` are particular important!
-- If you need to make your output more specific, change it (on the previous tab)...

## QUESTIONS RAISED BY THE LLM:

Let me address each point:

a) Key clarifying questions about the final output:

1. What is the intended final format of the samples? (WAV/MP3/etc, bit depth, sample rate)
2. What's the target number of samples in the final zip file? (The description mentions "48" but is this a hard requirement?)
3. Should the samples be normalized to a specific volume level?
4. Should the samples be labeled in any specific way? (e.g., "vocal_beat1_5s.wav" or "inst_4beat_130s.wav")
5. Should we maintain a consistent sample length across all extracts, or vary them?
6. Should we preserve any metadata about the original source (tempo, key, etc.)?
7. Do we need to ensure the samples are cleared of copyright?
8. Should there be a text file included listing the samples and their properties?

b) Required inputs before starting:

1. YouTube video URL
2. Three specific timestamp positions (in seconds)
3. Desired beat lengths for sample extraction (1/2/4/8 beats mentioned, but is this flexible?)
4. Target number of final samples
5. Whether to include both vocals and instrumentals
6. Any specific naming convention requirements
7. Output format specifications (bit depth, sample rate)

c) Task segmentation suggestions:

I see several natural segmentation points that could benefit from specialized handling:

1. Audio Acquisition Phase:
- YouTube download
- Audio quality verification
- Format standardization

2. Audio Analysis Phase:
- Tempo detection
- Beat grid alignment
- Key detection
- Audio quality assessment

3. Stem Separation Phase:
- Vocal isolation
- Instrumental isolation
- Quality verification of separation

4. Sample Generation Phase:
- Beat-aligned cutting
- Multiple length variants
- Normalization
- Quality checks

5. Sample Selection Phase:
- Amplitude analysis
- Frequency spectrum analysis
- Uniqueness checking
- Final filtering

Each of these could benefit from specialized handling and potentially different tools/approaches.

Expand

a) To understand the final output of the described task better, these clarifications may be helpful:

1) What is the specific format of the audio files expected? The user specified .wav, but it may be worth asking if other formats are acceptable or preferred.
   
2) Are there specific quality or bitrate requirements for the audio samples?
   
3) How long should the sections separating the instrumentals and vocals be for each of the chosen timeframes (5 seconds, 130 seconds, 190 seconds)? The user mentions 16 beats, but this could vary depending on the tempo of the song.
   
4) It was mentioned that the process may generate too many samples and a suggestion about filtering to the "best 48 samples" was made. On what criteria should the best samples be chosen?

5) The user mentioned saving files to a server and zipping them up. Would the user like these files to be saved in a specific directory or with a certain naming convention?
   
6) What exactly defines a 'beat'? Are we working with a 4/4 time signature or should the agent adapt to different musical signatures?

b) Before launching into the task completion process, this process should accept the following inputs:

1) The YouTube video from which the music samples would be extracted.
   
2) The specific timestamps in seconds for extracting the music samples (5 seconds, 130 seconds, 190 seconds).
   
3) Initial tempo/beat point for all six separate audio files.
   
4) The number of beats on which the extraction should begin.
   
5) The lengths and nature of the samples to be extracted (1/2/4/8 beat lengths).
   
6) Ideally, an idea about what user considers as 'the best' in terms of amplitude might allow better sample selection.

7) Finally, information about where to store and how to name the final .zip file, as well as the extracted .wav files will be required.

Research Links

Splice.com Vocal Chops Essential Collection 2023 WAV Samples Folder Structure Screenshot

Beatport Sounds Requirements Audio Sample Pack Submission Guidelines WAV Format 2023

Ableton Live 11 Sample Organization Tutorial YouTube Producer Workflow Screenshots

Loopmasters Hip Hop Vocal Chops Sample Pack WAV Files Components Review

GitHub Audio Sample Library Structure Python Stem Separation Project WAV Files

☝️ Like These? Consider Bringing These Ideas Into Output Or SubAgents (on previous page)

-- The LLM has suggested possible templates that could be useful to build this agent. 
-- Consider preparing these now...(you can upload them as `pre-prompts` in the next stage).
-- These prompts must be text, HTML or a URL (eg of an image or some folder).

## TEMPLATE IDEAS FROM THE LLM:

For this audio sampling task, here are valuable templates we could collect:

1. **Professional Sample Pack Listings**
- Visit Splice.com product pages for popular drum/vocal chop sample packs
- Screenshot and analyze the file organization/naming conventions of top-selling packs
- Review the technical specifications (sample rate, bit depth, file formats)
- Study how they organize and present their folder structures
- Example: https://splice.com/sounds/splice/vocal-chops-essential-collection

2. **Producer Marketplace Descriptions**
- Visit sites like Beatport Sounds, ADSR Sounds
- Analyze how professional sample packs are categorized and described
- Look at the technical requirements they mandate for submissions
- Example: https://www.beatport.com/sounds/requirements

3. **YouTube Producer Tutorials**
- Find videos where producers showcase their sample organization
- Look for screenshots/screencasts of their DAW folder structures
- Study how they name and categorize their samples
- Example: Search "how to organize samples FL Studio" or "Ableton sample management"

4. **GitHub Repositories**
- Search for open-source audio processing projects
- Study their output file naming conventions
- Review their folder structure specifications
- Example: https://github.com/topics/audio-samples

5. **Sample Pack Review Sites**
- Find professional reviews of commercial sample packs
- Study how reviewers describe organization and usability
- Note common praise/criticism about sample organization
- Example: https://producerhive.com/reviews/sample-packs/

These templates would help us refine:
- Precise naming conventions
- Optimal folder structures
- Industry-standard technical specifications
- Common organizational patterns
- Expected quantity/variety of samples

The goal would be to align our output with professional standards while maintaining practicality for our automated workflow.

Would you like me to dig deeper into any of these template sources or suggest additional ones?

Final Points Before Saving!

If you are at all unsure about this step, consider returning to step 1 (having taken new ideas from this Step 2 page)
DO NOT proceed to Step 3 if the information on this page has raised fundamental problems with your initial task
If the original description you wrote on Step 1 is largely correct, and this page has not really changed your mind about anything, then simply tweak the Output and SubAgents fields and Save (you can always hit BACK later)
IMPORTANT: If your initial description (Step 1) and Subagent & Output (Step 2) are flawless then the Agent will take it from here and your task will be easy to build. BUT if you are already seeing problems, consider hitting BACK and starting again.
It is MUCH easier to go back a step than to try to refine and imperfect task down the line. Final chance to start again!
Remember - the initial description on Step 1, along with the SubAgent summary & Output spec on Step 2, are the foundation on which all remaining steps and Agents run. This step is here for a reason, so please consider hitting back. It will save time later!

Need To Start Afresh?

BACK TO DESCRIBE

YouTube Audio Sample Generator - Refine SubAgent & Output

Describe

Refine

SubAgents

Deploy

Final Exact Output You Want ⚠️IMPORTANT⚠️

Expand

Other Possible SubAgents (Consider Adding These Above) ☝️🤔

Expand

Task Summary (You Wrote)

Expand

Diagram Of Entire Task (All SubAgents)

Expand Diagram

Ready To Proceed?

More Questions & Research Ideas (LLM #2)

Expand

Research Links

☝️ Like These? Consider Bringing These Ideas Into Output Or SubAgents (on previous page)

Final Points Before Saving!

Need To Start Afresh?

Tweaked & Good To Go?