<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:media="http://search.yahoo.com/mrss/">
<channel>
  <title>Vocaler Pulse: Lab</title>
  <link>https://pulse.vocaler.ai/lab/</link>
  <atom:link href="https://pulse.vocaler.ai/lab/feed.xml" rel="self" type="application/rss+xml"/>
  <description>Vocaler&#x27;s own prompt and audio experiments, with the settings.</description>
  <language>en</language>
  <image><url>https://pulse.vocaler.ai/og/home.png</url><title>Vocaler Pulse: Lab</title><link>https://pulse.vocaler.ai/lab/</link></image>
  <lastBuildDate>Fri, 13 Nov 2026 08:00:00 +0000</lastBuildDate>
  <item>
    <title>Lab: How well does Suno Bridge lock to an Ableton session?</title>
    <link>https://pulse.vocaler.ai/lab/lab-suno-bridge-tempo-key-lock-test/</link>
    <guid isPermaLink="true">https://pulse.vocaler.ai/lab/lab-suno-bridge-tempo-key-lock-test/</guid>
    <pubDate>Fri, 13 Nov 2026 08:00:00 +0000</pubDate>
    <category>Lab</category><category>suno bridge</category><category>ableton live</category><category>experiment</category>
    <description>A planned stress test of Bridge&#x27;s tempo, key and groove matching across odd meters, swing and key changes. The results shown are mock.</description>
    <content:encoded><![CDATA[<h2 id="hypothesis">Hypothesis</h2>
<p>Suno says Bridge output will "fit the tempo, key, style, harmonic content and groove" of a session. We expect near-perfect results on straight 4/4 pop, but drift on <strong>swing, odd meters and mid-loop key changes</strong>.</p>
<h2 id="setup">Setup</h2>
<ul><li><strong>Rig:</strong> Mac, Ableton Live 12, Suno Premier, Bridge beta. Each generation costs about 10 credits, so the test needs a budget of roughly 400 credits.</li><li><strong>Five test sessions</strong>, with an 8-bar loop window each:</li><li>120 BPM, 4/4, C minor house groove (control)</li><li>92 BPM hip-hop with 62% swing</li><li>7/8 math-rock riff at 140 BPM</li><li>6/8 folk ballad at 66 BPM</li><li>4/4 pop with a key change from G to A at bar 5 of the loop</li><li><strong>Prompt (identical for all sessions):</strong> <code>&quot;melodic bassline that locks to the kick, simple, supportive&quot;</code>, then a second pass with <code>&quot;airy backing vocal pad, oohs, no lyrics&quot;</code>.</li><li><strong>Eight takes per prompt per session</strong> (80 generations in total).</li><li><strong>Measures:</strong> onset offset from the grid (ms), key correctness (checked with a key detector and by ear), and a blind 1–5 "sits in the mix" score.</li></ul>
<h2 id="mock-findings-illustrative-only-not-real-results">Mock findings (illustrative only, not real results)</h2>
<div class="tw"><table><thead><tr><th>Session</th><th>Grid offset (median)</th><th>Key correct</th><th>Fit score</th></tr></thead><tbody><tr><td>House 4/4</td><td>6 ms</td><td>8/8</td><td>4.5</td></tr><tr><td>Swung hip-hop</td><td>18 ms</td><td>8/8</td><td>3.8</td></tr><tr><td>7/8 riff</td><td>35 ms</td><td>7/8</td><td>2.9</td></tr><tr><td>6/8 ballad</td><td>12 ms</td><td>8/8</td><td>4.0</td></tr><tr><td>Key change</td><td>9 ms</td><td>4/8</td><td>3.1</td></tr></tbody></table></div>
<p><em>Mock interpretation:</em> key and tempo hold up, but odd meters and modulations are the weak spots. Shorter loop windows that sit inside one key would be the practical fix.</p>
<h2 id="notes">Notes</h2>
<p>Use only sessions Vocaler fully owns, because Bridge uploads the project to Suno.</p>]]></content:encoded>
    <media:content url="https://pulse.vocaler.ai/og/lab-lab-suno-bridge-tempo-key-lock-test.png" medium="image" type="image/png" width="1200" height="630"/>
    <enclosure url="https://pulse.vocaler.ai/og/lab-lab-suno-bridge-tempo-key-lock-test.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>Lab: Can ElevenLabs match a video&#x27;s temp track with use_audio_as_reference?</title>
    <link>https://pulse.vocaler.ai/lab/lab-elevenlabs-video-to-music-audio-reference/</link>
    <guid isPermaLink="true">https://pulse.vocaler.ai/lab/lab-elevenlabs-video-to-music-audio-reference/</guid>
    <pubDate>Mon, 19 Oct 2026 08:00:00 +0000</pubDate>
    <category>Lab</category><category>elevenlabs</category><category>video</category><category>api</category><category>experiment</category>
    <description>A planned A/B test of ElevenLabs Video to Music with and without the audio-reference flag, checking BPM, genre and SFX timing. The results are mock.</description>
    <content:encoded><![CDATA[<h2 id="hypothesis">Hypothesis</h2>
<p>The <code>use_audio_as_reference</code> flag is meant to pull genre, BPM, instruments and mood from a video's own audio. With it on, we expect the generated score to land <strong>within ±5 BPM</strong> of the temp track and in the same genre far more often than with a text description alone.</p>
<h2 id="setup">Setup</h2>
<ul><li><strong>Clips:</strong> six 45-second edits, each with a royalty-free temp track Vocaler owns: lo-fi hip-hop (85 BPM), synthwave (110), orchestral trailer (70), drum &amp; bass (174), acoustic folk (100), afrobeats (105).</li><li><strong>Endpoint:</strong> <code>POST /v1/music/video-to-music</code>.</li><li><strong>Arm A:</strong> <code>description: &quot;score that matches the energy of the edit&quot;</code>, with <code>use_audio_as_reference: false</code>.</li><li><strong>Arm B:</strong> the same description, with <code>use_audio_as_reference: true</code>.</li><li><strong>Both arms:</strong> <code>add_sfx: true</code>, <code>separate_tracks: true</code>, and 3 generations per clip per arm (36 in total).</li><li><strong>Measures:</strong> BPM difference from the temp track, a blind genre-match vote from 3 listeners, SFX hits within ±100 ms of a visual event, and the number of SFX that had to be deleted.</li></ul>
<h2 id="mock-findings-illustrative-only-not-real-results">Mock findings (illustrative only, not real results)</h2>
<div class="tw"><table><thead><tr><th>Metric</th><th>Arm A (text only)</th><th>Arm B (audio reference)</th></tr></thead><tbody><tr><td>BPM within ±5</td><td>6/18</td><td>15/18</td></tr><tr><td>Genre match (majority vote)</td><td>9/18</td><td>16/18</td></tr><tr><td>SFX hits within ±100 ms</td><td>71%</td><td>73%</td></tr><tr><td>SFX deleted per clip (mean)</td><td>2.4</td><td>2.2</td></tr></tbody></table></div>
<p><em>Mock interpretation:</em> the reference flag mainly improves musical match. SFX timing doesn't change, which is expected, since effects are placed from the visuals.</p>
<h2 id="follow-ups">Follow-ups</h2>
<ul><li>Test drum &amp; bass on its own, because very high tempos may get halved.</li><li>Compare <code>separate_tracks</code> stems in a DAW mix against the combined output.</li></ul>]]></content:encoded>
    <media:content url="https://pulse.vocaler.ai/og/lab-lab-elevenlabs-video-to-music-audio-reference.png" medium="image" type="image/png" width="1200" height="630"/>
    <enclosure url="https://pulse.vocaler.ai/og/lab-lab-elevenlabs-video-to-music-audio-reference.png" type="image/png" length="0"/>
  </item>
  <item>
    <title>Lab: Does v6-wild really drift? A genre-adherence test across Suno v6 models</title>
    <link>https://pulse.vocaler.ai/lab/lab-suno-v6-vs-wild-vs-mini-genre-adherence/</link>
    <guid isPermaLink="true">https://pulse.vocaler.ai/lab/lab-suno-v6-vs-wild-vs-mini-genre-adherence/</guid>
    <pubDate>Wed, 09 Sep 2026 08:00:00 +0000</pubDate>
    <category>Lab</category><category>suno v6</category><category>experiment</category><category>prompting</category>
    <description>We plan to run one fixed brief through v6, v6-wild and v6-mini with Variety off to measure genre and tempo adherence. The findings below are mock.</description>
    <content:encoded><![CDATA[<h2 id="hypothesis">Hypothesis</h2>
<p>Suno describes <strong>v6-wild</strong> as pushing "further from the prompt." With <strong>Variety set to Off</strong>, we expect v6 to hit the requested genre and BPM most often, v6-wild least often, and v6-mini in between with thinner arrangements.</p>
<h2 id="setup">Setup</h2>
<ul><li><strong>Models:</strong> v6, v6-wild, v6-mini. <strong>10 generations each</strong> (30 total).</li><li><strong>Controls:</strong> Variety Off, Style Influence 50% (default), Weirdness default, instrumental off, same lyrics.</li><li><strong>Styles prompt:</strong></li></ul>
<div class="code"><button class="cp" type="button" data-copy>Copy</button><pre><code>1970s Ethiopian jazz, 96 BPM, minor pentatonic horn riffs, vibraphone,
fuzzy organ, tight funk drums, male baritone vocal in English, warm tape saturation
Exclude: EDM, trap hi-hats, autotune</code></pre></div>
<ul><li><strong>Lyrics:</strong> 2 verses and a chorus with <code>[Verse]</code> / <code>[Chorus]</code> tags.</li><li><strong>Scoring:</strong> BPM measured with a tempo detector (±3 BPM counts as a hit), blind genre-match votes from 3 listeners (1–5 scale), and whether any excluded element appears.</li></ul>
<h2 id="mock-findings-illustrative-only-not-real-results">Mock findings (illustrative only, not real results)</h2>
<div class="tw"><table><thead><tr><th>Metric</th><th>v6</th><th>v6-wild</th><th>v6-mini</th></tr></thead><tbody><tr><td>BPM within ±3</td><td>8/10</td><td>5/10</td><td>7/10</td></tr><tr><td>Mean genre-match score</td><td>4.1</td><td>3.2</td><td>3.6</td></tr><tr><td>Excluded element present</td><td>1/10</td><td>3/10</td><td>2/10</td></tr><tr><td>"Most interesting" listener pick</td><td>3</td><td>6</td><td>1</td></tr></tbody></table></div>
<p><em>Mock interpretation:</em> v6-wild loses adherence but wins on surprise, which supports the "scout with wild, finish with v6" workflow.</p>
<h2 id="follow-ups">Follow-ups</h2>
<ul><li>Repeat with Variety at Normal to measure how much prompt rewriting costs in adherence.</li><li>Test whether raising Style Influence closes v6-wild's gap.</li></ul>]]></content:encoded>
    <media:content url="https://pulse.vocaler.ai/og/lab-lab-suno-v6-vs-wild-vs-mini-genre-adherence.png" medium="image" type="image/png" width="1200" height="630"/>
    <enclosure url="https://pulse.vocaler.ai/og/lab-lab-suno-v6-vs-wild-vs-mini-genre-adherence.png" type="image/png" length="0"/>
  </item>
</channel>
</rss>
