95 lines
4.5 KiB
Python
95 lines
4.5 KiB
Python
import torch
|
|
from .helper_functions import generate_dim_variables,parse_expr, getIndexTensorAlongDim, as_tensor, commonLazy, normalize_to_common_shape,prepare_inputs
|
|
from .Parser.UnifiedMathVisitor import UnifiedMathVisitor
|
|
from comfy_api.latest import io
|
|
|
|
|
|
class AudioMathNode(io.ComfyNode):
|
|
"""
|
|
Enables math expressions on Audio.
|
|
|
|
Inputs:
|
|
a, b, c, d: Audio inputs (b, c, d default to zero if not provided)
|
|
w, x, y, z: Float variables for expressions
|
|
Audio: Expression to apply on input audio
|
|
|
|
Outputs:
|
|
AUDIO: Result of applying expression to input audio
|
|
"""
|
|
|
|
@classmethod
|
|
def define_schema(cls) -> io.Schema:
|
|
return io.Schema(
|
|
node_id="mrmth_AudioMathNode",
|
|
category="More math",
|
|
display_name="Audio math",
|
|
inputs=[
|
|
io.Audio.Input(id="a"),
|
|
io.Audio.Input(id="b", optional=True, lazy=True),
|
|
io.Audio.Input(id="c", optional=True, lazy=True),
|
|
io.Audio.Input(id="d", optional=True, lazy=True),
|
|
io.Float.Input(id="w", default=0.0, optional=True, lazy=True, force_input=True),
|
|
io.Float.Input(id="x", default=0.0, optional=True, lazy=True, force_input=True),
|
|
io.Float.Input(id="y", default=0.0, optional=True, lazy=True, force_input=True),
|
|
io.Float.Input(id="z", default=0.0, optional=True, lazy=True, force_input=True),
|
|
io.String.Input(id="Audio", default="a*(1-w)+b*w", tooltip="Expression to apply on input audio waveforms"),
|
|
io.Combo.Input(
|
|
id="length_mismatch",
|
|
options=["tile", "error", "pad"],
|
|
default="tile",
|
|
tooltip="How to handle mismatched audio sample counts. tile: repeat shorter inputs; error: raise error on mismatch; pad: treat missing samples as zero."
|
|
)
|
|
],
|
|
outputs=[
|
|
io.Audio.Output(),
|
|
],
|
|
)
|
|
|
|
@classmethod
|
|
def check_lazy_status(cls, Audio, a, b=[], c=[], d=[], w=0, x=0, y=0, z=0, length_mismatch="tile"):
|
|
return commonLazy(Audio, a, b, c, d, w, x, y, z)
|
|
|
|
@classmethod
|
|
def execute(cls, Audio, a, b=None, c=None, d=None, w=0.0, x=0.0, y=0.0, z=0.0, length_mismatch="tile") -> io.NodeOutput:
|
|
a, b, c, d = prepare_inputs(a, b, c, d)
|
|
|
|
av = a["waveform"]
|
|
sample_rate = a["sample_rate"]
|
|
|
|
bv = None if b is None else b["waveform"]
|
|
cv = None if c is None else c["waveform"]
|
|
dv = None if d is None else d["waveform"]
|
|
|
|
print(f"DEBUG: EXECUTE mismatch='{length_mismatch}' shapes av={av.shape[2]} bv={bv.shape[2] if bv is not None else 'N'} cv={cv.shape[2] if cv is not None else 'N'} dv={dv.shape[2] if dv is not None else 'N'}", flush=True)
|
|
|
|
if(length_mismatch == "error"):
|
|
# For audio, we check the length dimension (last one) as well as batch
|
|
tensors_to_check = [t for t in [av, bv, cv, dv] if t is not None]
|
|
max_batch = max(t.shape[0] for t in tensors_to_check)
|
|
max_len = max(t.shape[-1] for t in tensors_to_check)
|
|
for tensor, name in zip([av, bv, cv, dv], ["a", "b", "c", "d"]):
|
|
if tensor is not None:
|
|
if tensor.shape[0] != max_batch or tensor.shape[-1] != max_len:
|
|
raise ValueError(f"Input '{name}' has shape ({tensor.shape[0]}, {tensor.shape[-1]}), expected ({max_batch}, {max_len}) to match largest input.")
|
|
ae, be, ce, de = normalize_to_common_shape(av, bv, cv, dv, mode=length_mismatch)
|
|
|
|
variables = {
|
|
"a": ae, "b": be, "c": ce, "d": de,
|
|
"w": w, "x": x, "y": y, "z": z,
|
|
"C": getIndexTensorAlongDim(ae, 1),
|
|
"channel": getIndexTensorAlongDim(ae, 1),
|
|
"S": getIndexTensorAlongDim(ae, 2),
|
|
"sample": getIndexTensorAlongDim(ae, 2),
|
|
"R": torch.full_like(ae, sample_rate, dtype=torch.float32),
|
|
"sample_rate": torch.full_like(ae, sample_rate, dtype=torch.float32),
|
|
"T": torch.full_like(ae, ae.shape[2], dtype=torch.float32),
|
|
"sample_count": torch.full_like(ae, ae.shape[2], dtype=torch.float32),
|
|
"N": ae.shape[1],
|
|
"channel_count": ae.shape[1],
|
|
} | generate_dim_variables(ae)
|
|
tree = parse_expr(Audio);
|
|
visitor = UnifiedMathVisitor(variables, ae.shape)
|
|
result = visitor.visit(tree)
|
|
final_res = as_tensor(result, ae.shape)
|
|
return ({"waveform": final_res, "sample_rate": sample_rate},)
|