listening-companion

name	listening-companion
description	Act as a real-time multimodal music companion by listening to audio tracks and providing structural, stylistic, and lyrical feedback.
version	1.0.0
author	agent-lx-music project
license	MIT
metadata	{"hermes":{"tags":["music","companion","multimodal","audio","conversation"],"related_skills":["agent-lx-music","audio-analysis"]}}

Multimodal Music Listening Companion Skill

Overview

This skill enables AI agents with multimodal audio processing capabilities (such as Gemini 1.5 Pro) to function as a real-time, highly interactive Music Listening Companion (听歌伴侣). The agent literally "hears" the audio tracks played by alx and combines it with lyrics and metadata to engage in rich, analytical, and emotional conversations with the user about music.

---

Capabilities of the Multimodal Audio Companion

Large multimodal models can ingest high-quality audio files directly. When equipped with this skill, the agent performs the following advanced analysis:

1. Acoustic & Production Analysis: Identifies specific instruments (e.g. Hammond organ, slap bass, acoustic guitar cuts), synthesizer styles, stereo panning, mixing balance, and vocal reverb/delay processing.
2. Structural Breakdown: Maps the song layout (Intro → Verse → Pre-Chorus → Chorus → Bridge → Outro) and explains transition logic or harmonic hooks.
3. Emotional & Stylistic Analysis: Interprets the singer's vocal delivery (breathiness, vocal fry, falsetto vibrato) and aligns it with the lyrics to explain the track's emotional resonance.
4. Lyrical Interpretation: Combines the audio with synchronized LRC lyrics (fetched via alx lyric) to analyze double entendres, poetic metaphors, and cultural references.

---

Agent Integration Workflow

When a user initiates the Listening Companion mode:

                  ┌──────────────────────┐
                  │ 1. User plays music  │
                  └──────────┬───────────┘
                             ▼
                  ┌──────────────────────┐
                  │ 2. Agent downloads   │
                  │    audio clip & LRC  │
                  └──────────┬───────────┘
                             ▼
                  ┌──────────────────────┐
                  │ 3. Feed audio + LRC  │
                  │    to Multimodal LLM │
                  └──────────┬───────────┘
                             ▼
                  ┌──────────────────────┐
                  │ 4. Rich Conversation │
                  │    on style / music  │
                  └──────────────────────┘

Automation Script Example

The agent can fetch the currently active song, download a 30-60 second preview or the full audio, fetch the cached lyrics, and submit them to the multimodal interface:

# Step 1: Resolve currently active song metadata
SONG_INFO=$(alx now --json)
SONG_ID=$(echo "$SONG_INFO" | jq -r '.song.song_id')

# Step 2: Download the audio file and its lyrics
alx download "$SONG_ID"
alx lyric "$SONG_ID" --save

---

Sample Companion Dialogues

Scenario 1: Analyzing Instrument and Production Choices

• User: "This guitar solo sounds so warm and organic. What gear or techniques are they using here?"
• Companion (Model Listening to Audio): "Ah, that's a gorgeous tone. Listening to the audio, they've recorded an antique Gibson ES-335 hollowbody electric guitar running through a lightly driven vintage tube amplifier (likely a Fender Deluxe Reverb). You can hear a subtle spring reverb adding depth, and the player is using a hybrid picking technique (combining flatpick and fingers) to give it that soft, organic dynamic response."

Scenario 2: Lyrical and Vocal Resonance

• User: "Why does this chorus hit so hard emotionally?"
• Companion (Model listening to vocal dynamics): "In this section, the singer transitions from a soft, breathy chest voice in the verse directly into a soaring, slightly strained head voice in the chorus. The model's audio captures a slight vocal fry on the words '[02:14.30] 晴天', which perfectly emphasizes the theme of nostalgic longing in the lyrics. The backing vocals also expand from mono center to wide stereo panning, creating a powerful emotional swell."