laion/moss-va-sft3-dpo-lora

Preference-tuning adapter for MOSS voice-acting v2 SFT-3

A rank-64 LoRA trained with DPO on top of laion/moss-tts-local-transformer-4.55b-voice-acting-v2-sft3. This is the best checkpoint this project has measured, on the same 80-prompt generation-based evaluation used for everything else:

model	reward	WER	emotion pct	quality	burst	burst hit rate
SFT-3, no adapter	0.4584	0.0987	0.3494	0.9127	0.3564	0.666
+ DPO, corpus v1	0.4668	0.1117	0.3518	0.9108	0.3973	0.694
+ DPO, corpus v2 (step 4590)	0.4633	0.1024	0.3427	0.9062	0.3930	0.719
this adapter	0.4708	0.0950	0.3373	0.9235	0.4271	0.772

It is the first preference-tuned model in this line whose word error rate is better than the supervised baseline's (0.0950 against 0.0987) — every earlier DPO run paid for its gains in intelligibility.

What it was trained on

2,327,904 preference pairs, of which 20.5 % are a construction we call classifier-free-guidance pairs: two recordings of the same speaker within 10 % of each other's length, one strongly expressing a dimension and one not, with the spoken words removed from the prompt so the preference cannot be decided from the text. Each pair is emitted twice with the roles flipped — under a "high" instruction the intense clip is chosen, under a "low" instruction the mild one is — so a model cannot win by a global preference for loud audio and must condition on the instruction.

That worked: preference accuracy on those pairs rose from 0.562 (chance) to 0.975 during training.

What it did not do

It did not raise emotional intensity — 0.3373 against 0.3494 for the bare supervised model. The construction taught the model to attend to the instruction block and the timing tags, which is what the burst and WER numbers show, without giving it any new ability to reach an intensity it could not reach before. See the technical report for the three standing hypotheses and the one that was measured and refuted.

A note on checkpoint rotation

An earlier checkpoint of a different DPO run scored 0.4687 and was briefly the best; it was removed by the trainer's keep-last policy before it could be published. The adapter here is the best that still exists and the best ever measured.

Inference

import torch, torchaudio
from transformers import AutoProcessor, AutoModel

BASE = "laion/moss-tts-local-transformer-4.55b-voice-acting-v2-sft3"

proc = AutoProcessor.from_pretrained(BASE, trust_remote_code=True)
model = AutoModel.from_pretrained(BASE, trust_remote_code=True,
                                  dtype=torch.bfloat16, attn_implementation="sdpa").cuda().eval()

prompt = open("prompt.txt").read()          # the <user_inst> block, see "How to prompt" below
um = {"role": "user", "content": prompt, "audio_codes_list": []}
b = proc([[um]], mode="generation")

with torch.no_grad():
    out = model.generate(input_ids=b["input_ids"].cuda(),
                         attention_mask=b["attention_mask"].cuda(),
                         max_new_frames=340, do_sample=True,
                         audio_temperature=1.0, audio_top_p=0.95, audio_top_k=50,
                         audio_repetition_penalty=1.0)

# codes -> waveform. Use the processor's own decoder: calling the audio tokenizer directly, or
# reshaping its output, yields a two-channel result that flattens into audio at HALF SPEED and
# still sounds like speech. This project lost a whole corpus to that once.
wav = proc.decode_audio_codes([out_codes], return_stereo=False)[0].reshape(-1).float().cpu()
torchaudio.save("out.flac", wav[None], int(proc.model_config.sampling_rate), format="flac")

Loading adapters

from peft import PeftModel

# one adapter
model = PeftModel.from_pretrained(model, "laion/moss-va-sft3-dpo-lora")

# several, each with its own weight -- the usual case: identity from a voice adapter,
# affect from an emotion adapter, general quality from the DPO adapter.
#
# NOTE: `add_weighted_adapter(..., combination_type="linear")` does NOT work here. It raises
# `ValueError: All adapters must have the same r value`, because the DPO adapter is rank 64 and
# the voice / emotion adapters are rank 16. Activate them together instead and scale each one.
model = PeftModel.from_pretrained(model, "<dpo adapter path>", adapter_name="dpo")
model.load_adapter("<voice adapter path>", adapter_name="voice")
model.load_adapter("<emotion adapter path>", adapter_name="emo")

names = ["dpo", "voice", "emo"]
weights = {"dpo": 1.0, "voice": 1.0, "emo": 1.5}     # 1.5 for emotion is the measured optimum
model.base_model.set_adapter(names)                  # the TUNER takes a list; PeftModel does not
model.active_adapter = names[0]                      # must stay a str or generate() indexes a list

for mod in model.modules():
    sc = getattr(mod, "scaling", None)
    if isinstance(sc, dict):
        if not hasattr(mod, "_base_scaling"):
            mod._base_scaling = dict(sc)
        for k in sc:
            if k in weights:
                sc[k] = mod._base_scaling[k] * weights[k]

Scaling an adapter without re-merging

A LoRA layer computes h + scaling · B(A(x)), so multiplying the stored scaling is the merge weight — exact and reversible:

def set_lora_scale(model, w):
    for mod in model.modules():
        sc = getattr(mod, "scaling", None)
        if isinstance(sc, dict):
            if not hasattr(mod, "_base_scaling"):
                mod._base_scaling = dict(sc)
            for k in sc:
                sc[k] = mod._base_scaling[k] * w

How to prompt this model

Every request is one <user_inst> block. The fields are fixed — none may be added or removed:

<user_inst>
- Reference(s):
{None | Speaker: <name> | <|audio|>}
- Instruction:
{GENERAL: ... and/or SCRIPT: ...}
- Tokens:
{target length in audio frames}
- Quality:
None
- Sound Event:
None
- Ambient Sound:
None
- Language:
{English | German}
- Text:
{the same script as under SCRIPT:, character for character}
</user_inst>

Field	What goes in it
Reference(s)	<|audio|> when a reference recording of the target voice is attached, Speaker: <name> when only a voice name is known, otherwise None.
Instruction	A GENERAL: line, a SCRIPT: block, or both.
Tokens	Target length in audio frames. The tokenizer runs at 12.5 frames per second, so 12.8 s = 160 frames. This is the length budget and the numbers in the script must add up to it.
Quality, Sound Event, Ambient Sound	Always None. Kept so the field layout matches the base model.
Language	English or German.
Text	The rendered script, byte-identical to the SCRIPT: block.

GENERAL: — who is speaking

Prose describing the voice and the clip: age and gender, energy and pace, tension, timbre, clarity, pitch range, breath, affect, which emotions are audible, style, recording quality.

GENERAL: A young adult masculine voice; delivery is normally alert, brisk, neutral tension;
timbre is neutral-toned, fairly smooth; average clarity, wide pitch range, light breath;
affect is mildly positive, slightly dominant; reads as bitterness, contempt; 9.5s, EN.

The phrase reads as … is where the emotion names live.

SCRIPT: — what to say, when, and how

Four kinds of tag, told apart by their brackets:

Tag	Means	Rule
[3.9 seconds duration]	the next sentence must take this long	square brackets, stands before the text, one per speech segment
[0.8 seconds pause]	silence of this length	square brackets; every gap of 0.2 s or more, including before the first word and after the last
(contented sigh, 0.2 seconds)	a non-speech vocalisation of this length	round brackets with a duration — label first, then the seconds
(clearly amused, warm and open, unguarded)	how to perform the next sentence	round brackets without a duration, stands before the duration tag

The disambiguation rule in one line: square bracket = a number of seconds; round bracket with a number = a vocal burst; round bracket without a number = a delivery direction. That is the only thing separating a burst from a direction, which is why directions never carry a number.

A complete example:

<user_inst>
- Reference(s):
None
- Instruction:
GENERAL: A young adult feminine voice, warm and conversational; reads as amusement; 6.0s, EN.
SCRIPT:
[0.4 seconds pause] (intensely amused: letting it out / not hiding it, warm and open,
unguarded; bright, relaxed) [2.4 seconds duration] You are not going to believe this.
[0.3 seconds pause] (breathy giggle, 0.4 seconds) [0.2 seconds pause] (still intensely amused)
[2.3 seconds duration] He actually wore it to the wedding.
- Tokens:
75
- Quality:
None
- Sound Event:
None
- Ambient Sound:
None
- Language:
English
- Text:
[0.4 seconds pause] (intensely amused: letting it out / not hiding it, warm and open,
unguarded; bright, relaxed) [2.4 seconds duration] You are not going to believe this.
[0.3 seconds pause] (breathy giggle, 0.4 seconds) [0.2 seconds pause] (still intensely amused)
[2.3 seconds duration] He actually wore it to the wedding.
</user_inst>

0.4 + 2.4 + 0.3 + 0.4 + 0.2 + 2.3 = 6.0 s = 75 frames. If the numbers do not add up to the token budget the model has to choose which to honour, and length control is the thing it honours best.

Segmentation rules the training data followed

• split at sentence ends (. ! ? …) and at every vocal burst;
• any segment still longer than 12 s is split again at its largest internal gap;
• a duration is measured from the first word onset to the last word offset of that segment, so two sentences of 12 s and 8 s produce [12.0 seconds duration] and [8.0 seconds duration], never a single [20.0 seconds duration];
• gaps below 0.20 s are folded into the neighbouring speech instead of being printed, so the printed numbers still add up;
• a burst that overlaps speech prints only the part that does not overlap.

Vocal-burst labels that actually occur in training

low mumble, ahem, contented sigh, surprised gasp, chuckle, breathy giggle, childlike giggle, wistful sigh, exhausted groan, sharp inhale, resonant hum, scream, yawn, deep breath, soft hum, exasperated sigh, cackle, shriek, coughing, mournful wail, growl, purr.

Realistic lengths: median 0.28 s, 10th percentile 0.14 s, 90th percentile 0.48 s, longest observed 2.46 s. A sigh requested at 3 s is outside anything in the data.

Intensity bands

Delivery directions carry an intensity adverb drawn from the percentile band of the requested emotion. The same cutoffs are used by the training data, the reward and the evaluation:

band	percentile	adverbs
faint	0.40 – 0.70	barely, faintly, only slightly, just a little
moderate	0.70 – 0.90	clearly, plainly, noticeably, unmistakably
intense	0.90 – 0.98	strongly, intensely, very, deeply
extreme	0.98 – 1.00	overwhelmingly, extremely, utterly, completely

The family

🧩 Base model (required by every adapter here)	laion/moss-tts-local-transformer-4.55b-voice-acting-v2-sft3
🎚️ Preference-tuning adapter	laion/moss-va-sft3-dpo-lora
🎭 40 emotion adapters	laion/moss-va-sft3-emotion-loras
🗣️ 500 voice adapters	laion/moss-va-sft3-voice-loras
🔊 Listening page — nine models on the same 80 prompts, with ASR transcripts	laion/moss-va-sft3-samples
🔬 Emotion adapters vs baseline, matched and neutral prompts	laion/moss-va-emotion-loras
🧪 Four-factor study — what actually drives the emotion score	laion/moss-va-four-factor-study
📄 Technical report	laion/moss-va-technical-report
📖 Voice-acting manual (v2 model)	projects.laion.ai/moss-voiceacting-manual
⬅️ Predecessors	voice-acting-v2 · -sft · -sft-dpo